JPH11191135A - Japanese-English determination method of document image, document recognition method, and recording medium - Google Patents

Abstract

(57) [Summary] [PROBLEMS] To accurately and quickly identify Japanese and English, and to identify both in each character area and in each page. SOLUTION: After reducing an input document image 101, a black pixel connected component is extracted 103, and the extracted components are integrated to generate a character area 104. For the generated character area, the Japanese-English determination means 105 classifies the component based on the length of the connected component, and determines whether the area is a Japanese area or an English area based on the total value of the classification result.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method and a recording medium for determining whether a character region in a document image is a Japanese region or an English region, and to a method for determining whether each character region is a English region. The present invention relates to a document recognition method and a recording medium for performing recognition processing after determining whether an image is a Japanese document image or an English document image.

[0002]

2. Description of the Related Art When performing character recognition processing on a document image, it is necessary to select an appropriate language. In other words, even if you try to recognize Japanese by English OCR, you cannot recognize anything other than alphabets and numbers.
When attempting to recognize an English sentence using R, the recognition rate is lower than when using English sentence OCR after character extraction and language processing.

Therefore, it is necessary to perform language identification before performing the character recognition processing. 2. Description of the Related Art Various methods for identifying a character type in a document have conventionally been proposed. For example, there is a document recognition device that counts the number of black-and-white inversions in a vertical or horizontal direction of a binarized character line and identifies a character type based on the distribution (see Japanese Patent Application Laid-Open No. 5-108876).

There is also a document recognizing device that recognizes a read word and determines the language type of the recognized character based on the recognition result and the matching rate with the dictionary (see Japanese Patent Application Laid-Open No. 6-150061).

[0005]

In the former device, the number of black / white inversions is used as a feature for identifying a character type. This feature is based on fonts and document contents (kana, kanji,
(The ratio of numbers and the like) is large, which causes a problem that the accuracy of identification is reduced.

On the other hand, in the latter device, since character recognition is performed once, if the OCR performance is good, the character type can be determined with a considerable probability. Becomes possible. However, OCR has a problem that it takes a lot of time for processing.

SUMMARY OF THE INVENTION The present invention has been made in consideration of the above circumstances, and an object of the present invention is to accurately and quickly identify Japanese and English, and also determine the range of identification for each character area. In addition, a method and a recording medium for discriminating Japanese and English of a document image that can identify both for each page unit,
An object of the present invention is to provide a document recognition method and a recording medium that determine a document image and perform optimal document recognition processing.

[0008]

In order to achieve the above object, according to the first aspect of the present invention, it is determined whether each character area in a document image is a Japanese area or an English area. A Japanese English determination method, and determines whether the region is a Japanese region or an English region using a plurality of determination methods,
It is characterized in that a final judgment result is obtained by comparing the plurality of judgment results.

According to a second aspect of the present invention, there is provided a method for judging whether or not each character area in a document image is a Japanese area or an English area. The connected components are classified based on the lengths of the black pixel connected components in the character region generated by performing the above operation, and each of the character regions is a Japanese region or an English region based on a total value of the classification result. It is characterized by determining whether or not there is.

The invention according to claim 3 is characterized in that a different judgment method is used when the number of connected black pixel components in the generated character area does not satisfy a predetermined condition.

According to a fourth aspect of the present invention, there is provided a method for determining whether a document image on each page is a Japanese document image or an English document image, the method comprising reducing the size of the document image. Classifying the connected components based on the length of the black pixel connected components in the page generated by performing the above operation, and determining whether each of the pages is a Japanese region or an English region based on a total value of the classification result. Is determined.

According to the fifth aspect of the present invention, a page is composed of a plurality of character areas, and a Japanese / English determination method of a document image for determining whether a document image of each page is a Japanese document image or an English document image. Wherein the connected components are classified based on the lengths of the black pixel connected components in the character region generated by reducing the document image, and each of the character regions is classified based on the total value of the classification result. It is characterized in that it is determined whether the page is a Japanese area or an English area, and based on the determination result, it is determined whether each of the pages is a Japanese area or an English area.

According to a sixth aspect of the present invention, there is provided a method for judging whether a character area in a document image is a Japanese area or an English area. A line is detected, a block is extracted by integrating adjacent circumscribed rectangles from the line, and it is determined whether the block is a Japanese region, an English region, or an undeterminable region for each block. The judgment results are totaled for each block,
It is characterized in that it is determined whether each of the character areas is a Japanese area or an English area based on the total value.

According to a seventh aspect of the present invention, when the number of blocks to be extracted does not satisfy a predetermined condition, a different judgment method is used.

According to the invention described in claim 8, the page is composed of a plurality of character areas, and the Japanese / English determination method of the document image for determining whether the document image of each page is a Japanese document image or an English document image A line is detected from the character region, a circumscribed rectangle close to the line is integrated, a block is extracted, and it is impossible to determine whether the block is a Japanese region or an English region for each block. Determine whether the area
The determination results are tabulated on a page basis, and it is determined whether each page is a Japanese document image or an English document image based on the tabulated value.

According to the ninth aspect of the present invention, a page is composed of a plurality of character areas, and a Japanese / English determination method of a document image for determining whether the document image of each page is a Japanese document image or an English document image A line is detected from the character region, a circumscribed rectangle close to the line is integrated, a block is extracted, and it is impossible to determine whether the block is a Japanese region or an English region for each block. Determine whether the area
The determination result is totaled for each character region, and it is determined whether the region is a Japanese region or an English region for each character region based on the total value. The determination result is totaled for each page, and the total value is calculated. It is characterized in that it is determined whether each page is a Japanese document image or an English document image.

In the tenth aspect, it is determined whether the document image is a Japanese document image or an English document image.
It is characterized in that a document recognition process is performed according to the determination result.

According to the eleventh aspect, the document image is divided into a plurality of character areas, and it is determined whether each of the divided character areas is a Japanese document area or an English document area. It is characterized in that a document recognition process is performed in accordance with.

According to the twelfth aspect of the present invention, in order to determine whether each character region in a document image is a Japanese region or an English region, a plurality of determination methods are used to determine whether the character region is a Japanese region or an English region. It is a computer-readable recording medium that records a program for causing a computer to realize a function of determining whether an area is an area and a function of obtaining a final determination result by comparing the plurality of determination results. .

According to the thirteenth aspect, the document image is reduced in order to determine whether the document image of each character area or each page in the document image is a Japanese area or an English area. A function of classifying the connected component based on the length of the black pixel connected component in the generated character area or page, and each of the character areas or pages in the Japanese area based on the total value of the classification result. It is a computer-readable recording medium that records a program for causing a computer to realize a function of determining whether a region is an English language region.

According to the fourteenth aspect of the invention, in order to determine whether each character area in the document image is a Japanese area or an English area, or when a page is composed of a plurality of character areas, In order to determine whether the document image is a Japanese document image or an English document image, a function of detecting a line from the character area and a block that is extracted by integrating a circumscribed rectangle close to the line from the character region A function for determining whether each block is a Japanese area, an English area, or a non-determinable area; a function for totalizing the determination result for each block or for each page; Based on the value, a computer implements a function of determining whether each of the character areas is a Japanese area or an English area or a function of determining whether each page is a Japanese or English document image. Let It is characterized by a computer-readable recording medium recording a program for.

According to the fifteenth aspect of the present invention, the function of determining whether the document image is a Japanese document image or an English document image or dividing the document image into a plurality of character areas, A computer-readable recording medium that records a program for causing a computer to implement a function of determining whether a document area is a Japanese document area or an English document area and a function of performing document recognition processing according to the determination result It is characterized by having.

[0023]

DESCRIPTION OF THE PREFERRED EMBODIMENTS One embodiment of the present invention will be specifically described below with reference to the drawings. <Embodiment 1> FIG. 1 shows the structure of Embodiment 1 of the present invention.
In the figure, 101 is an image input unit for inputting a document image, 102 is an image reduction unit for reducing an input document image,
103 is a connected component extracting means for extracting connected components from the document image; 104 is a region generating means for generating a character region by classifying and integrating the extracted connected components;
5 is a Japanese-English discriminating means for discriminating between Japanese and English in a character area unit or a page unit, 106 is a control unit for controlling the whole, and 107 is various kinds of input document image data, connected component data, area data, etc. A data storage unit 108 for storing data is a data communication path, and 109 is a data communication unit connected to a host or the like via a network, a line, or the like.

FIG. 2 is a flowchart showing the entire process according to the first embodiment of the present invention. Hereinafter, the processing operation of the present invention will be described with reference to FIG. First, the image input unit 101
Obtains a document image by reading the document (step 201). This image input means is, for example, a scanner,
It may be a facsimile or the like, and an image may be obtained from another device via a network via the data communication means 109.

Next, the image reducing means 102 reduces the input document image (step 202). This process
For example, a process of OR-reducing the input document image to about 1/8. That is, the process is to reduce 8 × 8 pixels to one pixel, and if there is even one black pixel in 64 pixels, the reduced pixel is set to a black pixel.

The connected component extracting means 103 extracts a black pixel connected component from the reduced image (Step 203). The region generating means 104 classifies the extracted connected components and integrates them to generate a character region (step 204). As this area generation method, for example, a known method described in JP-A-6-20092 may be used. At this time, information on the connected components constituting each character area is stored and held in the data storage unit 107.

Subsequently, for the generated character area, the Japanese / English determining means 105 determines whether the character area is Japanese or English (step 205).

In step 202, the adjacent black pixels are merged by OR-reducing the image. Here, English sentences have the feature that there is a space between words and the space between characters in a word is very narrow. On the other hand, in Japanese, the character spacing does not change significantly except before and after punctuation.

FIG. 3 shows examples of images of English sentences and Japanese sentences and their circumscribed rectangles. The result obtained by reducing the English image 301 and extracting the connected components as a circumscribed rectangle is the circumscribed rectangle 30.
2 (note that the circumscribed rectangle 302
Should be smaller than the image 301, but are represented in the same size here). In an English image, a connected component is formed by fusing each word.

Regarding examples of Japanese images 303 and 305,
Similarly, when the connected component is extracted by being reduced, and is expressed by a circumscribed rectangle, the circumscribed rectangles 304 and 306 are obtained, respectively.

In the case of English sentences, since the number of characters constituting a word is constant to some extent, there is a feature that the circumscribed rectangle whose aspect ratio is about 2 to 6, or 7 is increased. On the other hand, in the case of Japanese, there is a feature that a long rectangle which is hard to appear in English sentence occurs as shown by a circumscribed rectangle 304 and, on the contrary, many small rectangles such as a circumscribed rectangle 306 occur.

Therefore, the above-described connected component rectangle is called “short”,
It is classified into three types of "medium" and "long", and these are totaled for each character area. FIG. 4 is a processing flowchart of the Japanese-English determination according to the first embodiment. The process of FIG. 4 is performed for each character area. When the row direction is horizontal, for example, the classification of rectangles is “short” when the width / height is 2 or less, “medium” when the width / height is 2 to 6, and “long” when the width / height is more than 2 (step). 401).
Then, the classification results in the character area are totaled (step 402), and it is determined whether the character area is Japanese or English (step 403). Here, the number of “short” rectangles is SC
NT, the number of “medium” rectangles is NCNT, and the number of “long” rectangles is L
Assuming CNT, the determination of Japanese or English is performed as shown in FIG. 8 (detailed flowchart of step 403).

First, LCNT / (NCNT + SCNT)
> Thl is checked (step 80)
1). Th1 is a predetermined threshold value.
It should be about 3. If this conditional expression holds, it means that there are many long rectangles, and it is determined that the character area is a Japanese area (step 804).

Next, when No is determined in step 801, it is checked whether NCNT / (LCNT + SCNT) <Th2 is satisfied (step 802). Th2 is also a predetermined threshold value, for example, 3. If this conditional expression holds, it means that there are few middle rectangles, and it is determined that the character area is a Japanese area (step 8).
04). If neither condition is satisfied, it is determined that the region is an English region (step 803).

<Embodiment 2> In the above-described embodiment 1, the judgment of Japanese or English is made for each character area. In this case, the number of characters may be very small depending on the character area. In such a case, the number of rectangles cannot be obtained sufficiently, so that it may be difficult to perform Japanese-English determination at the ratio of the number of rectangles.
The second embodiment is an embodiment that considers a case where the number of rectangles is not sufficient.

FIG. 5 shows a processing flowchart of the second embodiment. The Japanese-English discriminating means 105 checks whether or not the number of rectangles in the totaled area is sufficient (that is, whether or not the number is equal to or larger than a predetermined threshold Th) (step 501). Japanese-English discrimination using OCR described in JP-A-6-150061 is performed (step 5).
03). In this case, since the number of characters is small, even if OCR processing is performed, the increase in processing time is small. If the number of rectangles is sufficient, Japanese and English are identified by the rectangle length described in the first embodiment (step 502).

<Embodiment 3> Next, an embodiment 3 for performing Japanese-English discrimination on a page basis will be described. 6 and 7 show a detailed flowchart of step 205 according to the third embodiment.
The method shown in FIG. 6 counts the number of “short”, “medium”, and “long” rectangles not for each character area but for the entire page (steps 601 and 602), and uses the result to store the date in page units. An English determination is made (step 603). This Japanese / English determination method is performed according to the processing flowchart of FIG. The thresholds Th1 and Th2 at this time may be different from those in the case of processing in units of character areas.

In the method shown in FIG. 7, Japanese / English is determined for each character area (Step 702), and based on the result, Japanese / English is determined for the page (Step 703). Specifically, the number of regions determined to be Japanese regions is Jn, and the number of regions determined to be English regions is En, where Jn> En is a Japanese page, and En> Jn is an English page. J
If n = En, rejection may be performed, or the determination may be made in either Japanese or English.

<Embodiment 4> A description will be given of a Japanese-English identification method using features different from those of the above-described embodiment. FIG.
4 shows a configuration of a fourth embodiment. The difference from the first embodiment is that a line segmentation unit 902, a block extraction unit 903, and an in-block character type discrimination unit 904 are provided. Other components are the same as those of the first embodiment. FIG.
3 shows a processing flowchart.

First, the line cutout unit 902 cuts out a line from the character area of the document image (step 1001,
1002). Japanese Patent Application Laid-Open No. 6-2009
In the case of using the technology described in Japanese Patent Publication No. 2 (1993), line information is obtained at the stage of extracting a region, and this may be used. Segmentation of Document Image Using ”(Akiyama et al.,
August 1986, Vol. J69-D No. The method using projection described in 8) may be used.

Next, the block extracting unit 903 extracts a block corresponding to a word (step 1003). As the block extraction method, the present applicant has previously disclosed in Japanese Patent Application No. 8-347.
The method proposed in No. 81 may be used. That is, the block extracting unit 111 detects a circumscribed rectangle in the row data, and combines the circumscribed rectangle into block data. The method of combining the block data is as follows. A histogram of the character rectangle intervals (one rectangle is not yet determined to be one character. Therefore, in the case of kanji, a rectangle separated from bias and structure often becomes one rectangle) is obtained. FIG. 18 shows the extracted character rectangles and the distance between the rectangles. FIG. 19 shows a histogram of rectangular intervals.

In this histogram, the peak with the shortest distance tends to appear in the interval between the bias and formation of kanji and the distance between characters in the same proportional English word. Even if these are integrated, different character types rarely enter a block, so that block data is formed by integrating them. By performing this processing, proportional words and kanji characters in which one character is separated (that is, composed of bias and structure) are integrated into one.

The peak with the longest distance often appears at the distance between words and the distance between a punctuation mark and the next character.
These are often used at the boundary when the character type changes (especially the distance between words), and it is desired to avoid the same block. Therefore, processing is performed so that character rectangles having a distance equal to or longer than the longest peak value are not placed in the same block.

Further, the distance (A, B) between the target rectangle and the adjacent rectangles is measured. If the difference (A−B) is equal to or larger than a predetermined threshold, the rectangles having the longer distance are not integrated. Process to combine rectangles with shorter distances. FIG. 20 is a diagram illustrating a case where rectangles are not integrated at a position where the difference between the rectangles is large. In FIG. 20, three blocks are formed because rectangles are not integrated at a position where the difference is larger than a predetermined threshold. By performing such processing, even in a proportional English sentence or the like, even if the distance between words is absolutely short, the distance between characters must be different from the distance between characters, so that only one word can be integrated together. Even in a proportional font, Japanese kanji portions are arranged at relatively equal intervals, so that it is convenient to combine Japanese sentences.

By using the above-described block extraction method, in the case of an English sentence, unlike a Japanese document, words are separated by a space equivalent to a half-width, so that block data is mixed with other character types. That can be avoided.

Subsequently, a block character type discriminating unit 904 is used.
Performs a Japanese-English determination for each block (step 100).
4). This may also use the method of the above-mentioned application. In other words, the intra-block character type determination unit 904 determines whether the block grouped by the above process is a Japanese character or an alphanumeric character. The inside of the block is determined as the same character type. This character type is determined as follows. That is, when the number of black runs in the vertical direction or the number of black-and-white inversions of the rectangle in the block is equal to or greater than a predetermined threshold, the rectangle is identified as a Japanese character, and the rectangle in the extracted block in the vertical direction is Identify alphabetic characters based on coordinate values.
FIGS. 21A and 21B show specific examples of the number of vertical runs for Japanese and English characters. In an ideal case where there is no noise in alphanumeric characters, four runs can be made at maximum with the letter “g” (FIG. 21B). Therefore, when five or more runs are counted, the language is set to Japanese. In the case of the character "image" shown in FIG. 21A, the number of runs in the vertical direction changes as indicated by the number below the character.

The English / Japanese discriminating means 905 sums up the discrimination results for each block and performs Japanese / English discrimination of the area (step 1).
005). Here, the number of blocks determined as Japanese is JCNT, the number of blocks determined as English is ECNT,
The number of blocks determined to be undefined is defined as NCNT. FIG.
1 is a detailed flowchart of step 1005. If JCNT * Th3> ENCT, it is determined that the language is Japanese (steps 1101 and 1105).
When CNT> JCNT, it is determined to be English (1102,
1104). Otherwise, it is rejected (step 1103). The threshold value Th3 is, for example, 2.

<Embodiment 5> In Embodiment 4 described above, the judgment of Japanese or English is made in units of character areas. In this case, the number of characters may be very small depending on the character area. In such a case, the number of rectangles cannot be obtained sufficiently, so that it may be difficult to perform Japanese-English determination based on the ratio of the number of block determination results. Embodiment 5 is an embodiment in which the number of blocks is not sufficient.

FIG. 12 shows a processing flowchart of the fifth embodiment. The Japanese-English discriminating means 105 checks whether or not the total number of blocks in the character area is sufficient (that is, whether or not the number is equal to or greater than a predetermined threshold Th) (step 1201). Japanese-English distinction using OCR described in Japanese Patent Application Laid-Open No. Hei 6-150061 is performed (step 1203). In this case, since the number of characters is small, even if OCR processing is performed, the increase in processing time is small. Then, if the number of blocks is sufficient, Japanese and English are identified based on the determination result for each block described in the fourth embodiment (step 1202).

<Embodiment 6> In Embodiment 6, the Japanese / English distinction for each character area in Embodiment 4 is changed to Japanese / English distinction in page units. FIG. 6 is a processing flowchart of the sixth embodiment.
7 is used.

In the processing of FIG. 6, JCNT, ECN
The total of T and NCNT is calculated not for each character area but for the entire page, and the result is used to determine Japanese or English by the processing method of FIG. 11 described above. At this time, Th3 may be different from the case of the character area unit.

In the processing shown in FIG. 7, first, a determination is made for each character area, and the result is used to determine whether the page is Japanese or English. Specifically, the number of areas determined to be Japanese
n, the number of areas determined to be English areas as En, and Jn
If> En, it is determined to be a Japanese page, and if En> Jn, it is determined to be an English page. When Jn = En, rejection may be used, or any of Japanese and English may be used.

<Embodiment 7> In the embodiment 7, when performing the Japanese-English determination for each character area or for each page, as shown in FIG.
1) and Japanese-English determination processing (step 1302) using the determination result for each block, to determine Japanese and English, respectively. Then, from the respective determination results, a determination is finally made between Japanese and English (step 1303).

If both are determined to be Japanese or English, the final result may be determined to be Japanese or English as it is. If any are determined to be rejected,
The result of the determination that is not reject is the final result.

If the result of the determination is that one is Japanese and the other is English and the results do not match, one of the following is determined. (1) Reject. (2) The two confidence factors are calculated, and the result with the larger value is adopted. As the certainty factor of the discrimination method using the rectangular length, for example, LCNT / (NCNT + SCNT)> Thl, Thl
LCNT / (NCNT + SCNT) when = 0.3
* Value of 2.5 (upper limit is 1) NCNT / (LCNT + SCNT) <Th2, Th2
If = 3, (LCNT + SCNT) / NCNT *
NCNT / (LCNT + SCNT)> Th2 when the value of 2.5 (the upper limit is 1)
If N = 3, NCNT / (LCNT + SCNT) *
0.33 (upper limit is 1).

As the certainty factor of the discrimination method using the discrimination result for each block, for example, when JCNT * Th3> ECNT and Th3 = 2, the value of JCTN / (ECNT * 3) (the upper limit is 1)
If ECNT> JCNT, ECNT / JCNT *
0.7 (however, the upper limit is 1).

<Eighth Embodiment> FIG. 14 shows the structure of an eighth embodiment. FIG. 15 shows a processing flowchart of the eighth embodiment. In this embodiment, for the entire page of the input document, the Japanese-English discriminating unit 1412 uses the third embodiment described above.
Using the method of No. 6, Japanese-English discrimination processing is performed to determine whether the page is in Japanese or English (steps 1501 and 15).
02) Based on the determination result, the selection unit 1403 selects the English document recognition unit 1404 or the Japanese document recognition unit 1405, performs the document recognition processing of the selected language (steps 1504 and 1505), and outputs the recognition result. The data is output to an output unit such as a display (step 1506).

Since the attributes are different between Japanese and English, it may be better to switch the area division processing and the font identification processing. Therefore, the document recognition unit of this embodiment includes not only the character recognition processing but also the above-described area division processing and font identification processing.

<Embodiment 9> FIG. 16 shows the structure of the ninth embodiment, and FIG. 17 shows a processing flowchart of the ninth embodiment. The difference from the eighth embodiment is that Japanese-English identification is performed for each character area. For this purpose, the area dividing unit 1602 divides the input document into character areas (steps 1701 and 170).
2). Here, the region dividing unit uses a region dividing method applicable to both Japanese and English. After the division process, the Japanese / English discriminating unit 1603 performs a Japanese / English identification process for each character area using, for example, the method of the first embodiment described above (step 170).
4) Based on the discrimination result, the selecting unit 1604 selects the English document recognizing unit 1605 or the Japanese document recognizing unit 1606, performs a document recognizing process of the selected language (steps 1705 and 1706), and outputs the recognizing result. The output is output to an output unit 1607 such as a display (step 1707).
The document recognition unit according to the ninth embodiment performs a font identification process in addition to the document recognition process.

<Embodiment 10> In each of the above embodiments, Japanese and English are determined using black pixel connected components and rectangular lengths as feature amounts. However, the determination method using the black pixel connected component takes a long processing time, and the method using the rectangular length may cause high rejection. There is also a method of distinguishing between Japanese and English based on the peak position of the frequency distribution of the relative position in the upper and lower lines of the circumscribed rectangle (Japanese Patent Publication No.
However, when a document having an inclination is input, there is a problem that the frequency distribution greatly changes and the identification accuracy is reduced.

Therefore, in the present embodiment, the line height is
By distinguishing between Japanese and English using the histogram of the height of the circumscribed rectangle in the line, Japanese and English are accurately and quickly identified for each area of the document image. Then, for an area that cannot be determined by the above-described Japanese-English identification method, Japanese-English identification is performed using another method.

FIG. 22 shows the structure of the tenth embodiment. FIG. 23 is an overall processing flowchart of the tenth embodiment. First, the image input unit 2201 obtains a document image by reading a document (step 2301).
The image input unit is, for example, a scanner, a facsimile, or the like, and an image may be obtained from another device via the data communication unit 2207 via a network.

Next, the area generating means 2202 generates a character area (step 2302). For example, a method described in JP-A-6-20092 may be used as the region generation method. Next, the line segmentation means 2203
Extracts a line for character recognition from a character area. That is, the circumscribed rectangles of the characters are obtained, and they are integrated to generate a line (step 2303). Japanese-English identification means 22
04 performs Japanese-English identification for the generated character area (step 2304).

The discrimination between Japanese and English is performed as follows. FIG.
FIG. 7 is a detailed flowchart of the Japanese-English identification (step 2304). FIG. 24 shows an example of a cut-out line and a circumscribed rectangle in the line. First, the frequency distribution of the ratio of the height of the circumscribed rectangle in the row to the row height is calculated (step 270).
1, 2702). The line height is lineheight, and the rectangle height is height. Weight ratio
Let e = height * 100 / lineheight. In the case of a document having a slope as shown in FIG. 25, the maximum value of the height of the rectangle of the line may be used as lineheight instead of the line height in order to more accurately identify Japanese and English. That is, with respect to an input document having a slope, Japanese and English are identified based on a histogram of the ratio of the height of each circumscribed rectangle in the line to the maximum height of the rectangle in the line.

If the above-mentioned ratio heightrate is, for example, 80 or more, the number of rectangles is set to lcnt, and
The number of rectangles when the rate is, for example, 70 or more and less than 80 is ncnt, and the number of rectangles when the heightrate is, for example, 40 or more and less than 70 is scnt. Lcnt, ncnt, scnt for all rectangles in the character area
Ask for.

FIG. 26 shows an example of the number of rectangles examined for a Japanese document and an English document. In general, Japanese is lcnt
English tends to have a large scnt.
Therefore, predetermined thresholds thJ and thE are set, and lcnt
If / scnt> thJ, it is determined that the language is Japanese (step 2
703), if lcnt / scnt <thE, it is determined that the language is English (step 2704). Otherwise, it is set as an unknown area (step 2705).

The above-mentioned unknown area can be distinguished between English and Japanese by using a statistical method. FIG. 28 is a detailed processing flowchart for an unknown area. For example, feature values lcnt, nc of the Japanese region and the English region in advance
nt and scnt are normalized, and the mean and the inverse matrix of the covariance matrix are obtained for Japanese and English, respectively. Then, the Mahalanobis distance is calculated for each of Japanese and English using the average value and the inverse matrix of the covariance matrix (steps 2801 and 2802).

When the Mahalanobis distance in Japanese is Dj and the Mahalanobis distance in English is De, the predetermined threshold is M
Assuming e and Mj, English is determined when Dj / De> Me (step 2803), and Japanese is determined when Dj / De <Mj (step 2804). If none of the conditions is satisfied, it is determined that the area is unknown (step 280).
5). Instead of the Mahalanobis distance described above, a Euclidean distance with an average value or a city block distance may be used.

Further, for an area determined to be unknown, Japanese-English discrimination is performed using the certainty factor of English sentence recognition. FIG. 29 is a detailed processing flowchart of step 2805. The confidence is calculated by English sentence recognition (step 2901). Next, for the calculated certainty, for example, the number of words having a certainty of 60% or more is Good, the number of words having a certainty of less than 60% is 0, and the number of words having a certainty of 0 is Zero. (Step 2902).

When the judgment value of Japanese-English discrimination is set to Value, Value = Good / (Good + Bad + Ze
lo) (step 2903), and Value is a predetermined threshold value t.
If it exceeds heocr (step 2904), it is determined that the language is English, and if it is less than it, it is determined that it is Japanese.

The weight may be assigned to Zero. Z
If ero is, for example, three Bads, Value
Since Bad = Bad + Zero × 3, Value = Good / (Good + Bad), and if the value exceeds the threshold theeocr, it can be determined that the language is English, and if it is less than that, the language can be determined to be Japanese. In this way, even in an area where the number of characters for Japanese / English identification is small, the English / Japanese identification is performed with the certainty based on the English sentence recognition.

<Embodiment 11> In this embodiment, a circumscribed rectangle is generated from an image obtained by reducing an input document image, an appropriate integration is performed between the generated rectangles, and a histogram of the rectangle length aspect ratio after the integration is obtained. This is an embodiment in which Japanese-English discrimination is performed with higher accuracy by using this method.

FIG. 30 shows the structure of the eleventh embodiment. FIG. 31 is an overall processing flowchart of the eleventh embodiment. Image input means 30 in the same manner as in the above-described embodiment.
01 is input to the image reducing unit 30.
02 (steps 3101 and 310
2). In this processing, for example, the document image is ORed to about 1/4.
Compression (4 × 4 pixels are reduced to 1 pixel, and if there is at least one black pixel in 16 pixels, the reduced image is black).

Next, the area generating means 3003 generates a character area (step 3103). For example, a method described in JP-A-6-20092 may be used as the region generation method. Subsequently, the rectangle integration means 3004
Performs integration of rectangles so that the characteristics of Japanese and English can be well expressed (step 3104). For example, as shown in FIG. 32, when the vertical coordinates of the y-coordinates (vertical direction) of the rectangles 1 and 2 are close and the x-coordinates of the adjacent rectangles 1 and 2 are very close (for example, the horizontal distance between the rectangles). Is smaller than the distance corresponding to the English space), the rectangle is integrated. Also, for example,
As shown in FIG. 33, the left rectangle 1 is replaced with the right rectangle 2 by y.
The rectangles 1, which are in a positional relationship encompassed by coordinates and are adjacent to each other,
2 is very close (eg, the horizontal distance between rectangles is less than the distance corresponding to English space)
Merge rectangles.

Then, using the rectangular aspect ratio (rectangular long vertical / rectangular long horizontal), four rectangles of a long rectangle, a medium rectangle, a small rectangle, and a very small rectangle are obtained.
It is divided into two features (FIG. 34). In general, Japanese has a high proportion of long rectangles, and English has a high proportion of medium rectangles. Utilizing this difference in characteristics, the English-Japanese identification means 3005 creates an identification determination formula and performs Japanese-English identification (step 3105). FIG. 35 is a detailed flowchart of the Japanese-English identification processing.

For example, the number of long rectangular areas in the area lcnt The number of medium rectangular areas in the ncnt area The number of small rectangular areas in the ncnt area The number of small rectangular areas in the scnt area The number of small rectangular areas sscnt (in the case of noise, Is calculated (step 3501), and the ratio of the long rectangle in the area ratio1 = lcnt / (ncnt + sc)
nt) is calculated (step 3502), and the ratio of the middle rectangle in the area ratio2 = ncnt / (lcnt + scn)
t) is calculated (step 3503). When calculating the above ratio, sscnt was ignored as noise.

Then, ratio is defined as x coordinate, ratio
Let o2 be the y coordinate, erroneous identification is reduced as much as possible, and the overlapped portion is divided into a Japanese region, an English region, and a reject region so as to be rejected. For example, ratio2
If / ratio> thE, it is determined that the region is an English region (step 3504), and ratio2 / ratio <thJ
If so, it is determined that the area is a Japanese area (step 3505), and the other areas are unknown in Japanese and English (step 3506).
Here, thE and thJ are predetermined thresholds.

In the same manner as in the tenth embodiment, the area determined to be Japanese / English unknown is identified using a statistical method. For example, the feature value lcn of the Japanese region and the English region in advance
The t, ncnt, and scnt are normalized, and the average value and the inverse matrix of the covariance matrix are obtained in Japanese and English, respectively. The Mahalanobis distance for each of Japanese and English is calculated using the mean and the inverse matrix of the covariance matrix. When the Mahalanobis distance in Japanese is Dj and the Mahalanobis distance in English is De, and given thresholds are Me and Mj, Dj / De> M
If e, English is determined, and if Dj / De <Mj, Japanese is determined. If none of the conditions is satisfied, it is determined to be unknown. Note that, instead of the Mahalanobis distance, a Euclidean distance with an average value or a city block distance may be used.

<Embodiment 12> The present invention is not limited to the above-described embodiment, and can be realized by software. When the present invention is implemented by software, as shown in FIG. 36, a CPU, a memory, a display device,
Hard disk, keyboard, CD-ROM drive,
Prepare a computer system consisting of a scanner, etc.
On a computer-readable recording medium such as a CD-ROM, a program for realizing the Japanese-English determination function and the document recognition function of the present invention is recorded. A document image or the like input from an image input unit such as a scanner is temporarily stored in a hard disk or the like. Then, when the program is started, the temporarily stored document image data is read, a Japanese-English determination process and a document recognition process are executed, and the results are output to a display or the like.

[0080]

As described above, claims 1 and 12
According to the described invention, a plurality of determination methods are used in combination, so that Japanese and English can be determined with high accuracy.

According to the second, third, sixth, seventh, thirteenth, and fourteenth aspects, it is possible to accurately determine Japanese and English for each character area in a document image.

According to the fourth, fifth, eighth, ninth, thirteenth, and fourteenth aspects of the invention, it is possible to accurately determine Japanese and English for each page of a document image.

According to the tenth, fifteenth, and fifteenth aspects, a document image determined to be Japanese or English is
Since an appropriate document recognition process is performed, a highly accurate recognition result can be obtained.

[Brief description of the drawings]

FIG. 1 shows a configuration of a first exemplary embodiment of the present invention.

FIG. 2 shows an overall processing flowchart of Embodiment 1 of the present invention.

FIG. 3 shows image examples of English sentences and Japanese sentences and their circumscribed rectangles.

FIG. 4 is a processing flowchart of Japanese-English determination according to the first embodiment.

FIG. 5 shows a processing flowchart of a second embodiment.

FIG. 6 shows a first detailed flowchart of step 205 according to the third embodiment.

FIG. 7 shows a second detailed flowchart of step 205 according to the third embodiment.

FIG. 8 shows a detailed flowchart of step 403.

FIG. 9 shows a configuration of a fourth embodiment.

FIG. 10 shows a processing flowchart of a fourth embodiment.

FIG. 11 is a detailed flowchart of step 1005.

FIG. 12 shows a processing flowchart of a fifth embodiment.

FIG. 13 shows a processing flowchart of a seventh embodiment.

FIG. 14 shows a configuration of an eighth embodiment.

FIG. 15 shows a processing flowchart of the eighth embodiment.

FIG. 16 shows a configuration of a ninth embodiment.

FIG. 17 shows a processing flowchart of the ninth embodiment.

FIG. 18 shows an extracted character rectangle and a distance between the rectangles.

FIG. 19 shows a histogram of rectangular intervals.

FIG. 20 is a diagram illustrating a case where rectangles are not integrated at a position where the difference between the rectangles is large.

FIGS. 21A and 21B show specific examples of the number of vertical runs for Japanese and English characters.

FIG. 22 shows a configuration of a tenth embodiment.

FIG. 23 is an overall processing flowchart of the tenth embodiment.

FIG. 24 shows an example of a cut-out line and a circumscribed rectangle in the line.

FIG. 25 shows an example of a line when a document is inclined and a circumscribed rectangle in the line.

FIG. 26 shows an example of the number of rectangles examined for a Japanese document and an English document.

FIG. 27 is a detailed processing flowchart of Japanese-English identification (step 2304).

FIG. 28 is a detailed processing flowchart for an unknown area.

FIG. 29 is a detailed processing flowchart of step 2805.

FIG. 30 shows a configuration of an eleventh embodiment.

FIG. 31 is an overall processing flowchart of an eleventh embodiment.

FIG. 32 shows an example of integrating rectangles.

FIG. 33 shows another example of merging rectangles.

FIG. 34 shows rectangles classified into four types.

FIG. 35 is a detailed processing flowchart of Japanese-English identification processing in Example 11;

FIG. 36 shows a structure of a twelfth embodiment.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 101 Image input means 102 Image reduction means 103 Connected component extraction means 104 Area generation means 105 Japanese-English discrimination means 106 Control unit 107 Data storage unit 108 Data communication path 109 Data communication means

────────────────────────────────────────────────── ───

[Procedure amendment]

[Submission date] July 15, 1998

[Procedure amendment 1]

[Document name to be amended] Statement

[Correction target item name] Full text

[Correction method] Change

[Correction contents]

[Document Name] Statement

Patent application title: Method for judging Japanese and English in document image, document recognition method and recording medium

[Claims]

10. A method for determining whether a character region in a document image is a Japanese region or an English region, the method comprising the steps of: cutting out a line from each character region; A Japanese-English determination method for a document image, comprising determining whether each of the character regions is a Japanese region or an English region based on a height and a rectangular height in a line.

11. height and row rectangle height of the line, Japanese English document image according to claim 10 which is a histogram of the percentage of row each rectangle height to the height of the row Judgment method.

12. A method for judging whether a character region in a document image is a Japanese region or an English region, the method comprising the steps of: cutting out a line from each character region; Based on the histogram of the ratio of the height of each rectangle in the line to the height, it is determined whether each of the character areas is a Japanese area or an English area. For an unknown area that cannot be determined in any area, The characteristic value of Japanese and the characteristic value of English are calculated based on the histogram, and it is determined whether the unknown region is the Japanese region or the English region by calculating which characteristic value is closer to the unknown region. A method for determining Japanese / English of a document image, characterized by determining.

13. Each character area in the document image is a Japanese English determination method of determining a document image whether the English area or a Japanese area, wherein from each of the character areas cut out line, row A Japanese-English determination method for a document image, comprising determining whether each of the character areas is a Japanese area or an English area based on a maximum height of a rectangle and a height of a rectangle in a line.

14. The method according to claim 13, wherein the maximum height of the rectangle in the row and the height of the rectangle in the row are histograms of the ratio of the height of each rectangle in the row to the maximum height of the rectangle in the row. Japanese / English judgment method for document images.

15. Each character area in the document image is a Japanese English determination method of determining a document image whether the English area or a Japanese area, wherein from each of the character areas cut out line, row Based on the histogram of the ratio of the height of each rectangle in the line to the maximum height of the rectangle, it is determined whether each of the character regions is a Japanese region or an English region. In advance, a characteristic value of Japanese and a characteristic value of English are calculated based on the histogram, and by calculating which characteristic value the unknown region is closer to the characteristic value, it is determined whether the region is the Japanese region or the English region. A method for judging Japanese or English of a document image, characterized by judging whether or not there is a document image.

16. An unknown area that cannot be determined even in the second determination is determined as to whether it is a Japanese area or an English area based on the confidence of English sentence recognition. Japanese / English judgment method of the document image described.

17. A method for determining whether each character region in a document image is a Japanese region or an English region, the method comprising: determining a circumscribed rectangle from a reduced image of the document image; Generating, integrating rectangles having a predetermined positional relationship, and determining whether each of the character regions is a Japanese region or an English region based on a histogram of the aspect ratio of the rectangle after the integration. Japanese / English judgment method of the document image to be executed.

18. Each character area in the document image is a Japanese English determination method of determining a document image whether the English area or a Japanese region, circumscribed rectangles from the image obtained by reducing the document image Generate and integrate rectangles having a predetermined positional relationship, and determine whether each of the character regions is a Japanese region or an English region based on a histogram of the aspect ratio of the rectangle after the integration. For an unknown area that cannot be determined even in advance, the Japanese characteristic value and the English characteristic value are calculated in advance based on the histogram, and by calculating which characteristic value the unknown area is closer to, A Japanese-English determination method for a document image, which determines whether the image is a word area or an English area.

25. A function for cutting out a line from each character area in a document image, a histogram of the ratio of the height of each rectangle in the line to the height of the line, or the height of each rectangle in the line relative to the maximum height of a rectangle in the line. A function of determining whether each of the character areas is a Japanese area or an English area based on the histogram of the ratio; and an unknown area that cannot be determined as any area, the characteristic of Japanese based on the histogram in advance. The computer realizes a function of calculating a value and a characteristic value of English, and a function of determining whether the unknown region is closer to the characteristic value to determine whether the region is a Japanese region or an English region. Readable recording medium on which a program for recording is recorded.

And 26. A function of generating a reduced image from the document image, a function of generating a character area from the reduced image, a function of generating an enclosing rectangle from the character area, a rectangle with each other in a predetermined positional relationship integrating And a computer-readable program storing a program for causing a computer to implement a function of determining whether the character area is a Japanese area or an English area based on a histogram of an aspect ratio of a rectangle after integration. recoding media.

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method and a recording medium for determining whether a character region in a document image is a Japanese region or an English region, and to a method for determining whether each character region is a English region. The present invention relates to a document recognition method and a recording medium for performing recognition processing after determining whether an image is a Japanese document image or an English document image.

[0002]

2. Description of the Related Art When performing character recognition processing on a document image, it is necessary to select an appropriate language. In other words, even if you try to recognize Japanese by English OCR, you cannot recognize anything other than alphabets and numbers.
When attempting to recognize an English sentence using R, the recognition rate is lower than when using English sentence OCR after character extraction and language processing.

Therefore, it is necessary to perform language identification before performing the character recognition processing. 2. Description of the Related Art Various methods for identifying a character type in a document have conventionally been proposed. For example, there is a document recognition device that counts the number of black-and-white reversals in a vertical or horizontal direction of a binarized character line and identifies a character type based on the distribution (see Japanese Patent Application Laid-Open No. 5-108876).

There is also a document recognizing device that recognizes a read word and determines the language type of the recognized character based on the recognition result and the matching rate with the dictionary (see Japanese Patent Application Laid-Open No. 6-150061).

[0005]

In the former device, the number of black / white inversions is used as a feature for identifying a character type. This feature is based on fonts and document contents (kana, kanji,
(The ratio of numbers and the like) is large, which causes a problem that the accuracy of identification is reduced.

On the other hand, in the latter device, since character recognition is performed once, if the OCR performance is good, the character type can be determined with a considerable probability. Becomes possible. However, OCR has a problem that it takes a lot of time for processing.

The present invention has been made in consideration of the above circumstances, and an object of the present invention is to accurately and quickly identify Japanese and English, and also to determine the range of identification for each character area. In addition, a method and a recording medium for discriminating Japanese and English of a document image that can identify both for each page unit,
An object of the present invention is to provide a document recognition method and a recording medium that determine a document image and perform optimal document recognition processing.

[0008]

In order to achieve the above object, according to the first aspect of the present invention, it is determined whether each character area in a document image is a Japanese area or an English area. A Japanese English determination method, and determines whether the region is a Japanese region or an English region using a plurality of determination methods,
It is characterized in that a final judgment result is obtained by comparing the plurality of judgment results.

According to a second aspect of the present invention, there is provided a method for judging whether or not each character area in a document image is a Japanese area or an English area. The connected components are classified based on the lengths of the black pixel connected components in the character region generated by performing the above operation, and each of the character regions is a Japanese region or an English region based on a total value of the classification result. It is characterized by determining whether or not there is.

The invention according to claim 3 is characterized in that a different judgment method is used when the number of connected black pixel components in the generated character area does not satisfy a predetermined condition.

According to a fourth aspect of the present invention, there is provided a method for determining whether a document image on each page is a Japanese document image or an English document image, the method comprising reducing the size of the document image. Classifying the connected components based on the length of the black pixel connected components in the page generated by performing the above operation, and determining whether each of the pages is a Japanese region or an English region based on a total value of the classification result. Is determined.

According to the fifth aspect of the present invention, a page is composed of a plurality of character areas, and a Japanese / English determination method of a document image for determining whether a document image of each page is a Japanese document image or an English document image. Wherein the connected components are classified based on the lengths of the black pixel connected components in the character region generated by reducing the document image, and each of the character regions is classified based on the total value of the classification result. It is characterized in that it is determined whether the page is a Japanese area or an English area, and based on the determination result, it is determined whether each of the pages is a Japanese area or an English area.

According to a sixth aspect of the present invention, there is provided a method for judging whether a character area in a document image is a Japanese area or an English area. A line is detected, a circumscribed rectangle close to the line is integrated, a block is extracted, and it is determined whether the block is a Japanese region, an English region, or an undeterminable region for each block. The judgment results are totaled for each block,
It is characterized in that it is determined whether each of the character areas is a Japanese area or an English area based on the total value.

According to a seventh aspect of the present invention, when the number of blocks to be extracted does not satisfy a predetermined condition, a different judgment method is used.

According to the invention described in claim 8, the page is composed of a plurality of character areas, and the Japanese / English determination method of the document image for determining whether the document image of each page is a Japanese document image or an English document image A line is detected from the character region, a circumscribed rectangle close to the line is integrated, a block is extracted, and it is impossible to determine whether the block is a Japanese region or an English region for each block. Determine whether the area
The determination results are tabulated on a page basis, and it is determined whether each page is a Japanese document image or an English document image based on the tabulated value.

According to the ninth aspect of the present invention, a page is composed of a plurality of character areas, and a Japanese / English determination method of a document image for determining whether the document image of each page is a Japanese document image or an English document image A line is detected from the character region, a circumscribed rectangle close to the line is integrated, a block is extracted, and it is impossible to determine whether the block is a Japanese region or an English region for each block. Determine whether the area
The determination result is totaled for each character region, and it is determined whether the region is a Japanese region or an English region for each character region based on the total value. The determination result is totaled for each page, and the total value is calculated. It is characterized in that it is determined whether each page is a Japanese document image or an English document image.

According to the tenth aspect of the present invention, in the document image,
Determines whether each character area is a Japanese area or an English area.
A method of determining Japanese / English of a document image to be determined,
Cut line from character area, line height and rectangle height in line
Based on the above, each character area is a Japanese area or an English area
It is characterized by determining whether or not there is.

In the invention according to claim 11, the height of the row is
And the rectangle height in the row is the height of each rectangle in the row relative to the row height
It is characterized in that it is a histogram of the ratio of the height.

According to the twelfth aspect of the present invention, in the document image,
Determines whether each character area is a Japanese area or an English area.
A method of determining Japanese / English of a document image to be determined,
Cut out the line from the character area and set each line in the line to the line height.
Based on the histogram of the rectangular height ratio,
Judge whether the area is a Japanese area or an English area.
For unknown regions that cannot be determined as regions,
Calculates Japanese characteristic values and English characteristic values based on the strogram
Which of the characteristic values the unknown area is close to
By calculating
It is characterized by determining whether or not there is.

According to the thirteenth aspect, in the document image,
Determines whether each character area is a Japanese area or an English area.
A method of determining Japanese / English of a document image to be determined,
Cut a line from the character area, and the maximum height and line of the rectangle in the line
Each character area is a Japanese area based on the height of the rectangle inside
It is characterized in that it is determined whether the region is an English region.

According to the present invention, the rectangular shape in the line is
The maximum height of the shape and the height of the rectangle in the row are the maximum height of the rectangle in the row
Histogram of the ratio of the height of each rectangle in the row to the height
It is characterized by that.

According to the fifteenth aspect, the document image includes
Determines whether each character area is a Japanese area or an English area.
A method of determining Japanese / English of a document image to be determined,
Cut out the line from the character area, and match the maximum height of the rectangle in the line.
Based on the histogram of the height ratio of each rectangle in the row
Determines whether each character area is a Japanese area or an English area.
Unknown area that cannot be determined in any area
Is based on the histogram,
And the characteristic value of
By calculating whether or not it is close to the characteristic value,
Or English language domain
You.

In the invention according to claim 16, the second time
For unknown areas that cannot be determined by judgment,
Based on the certainty factor, it is determined whether the language is Japanese or English.
It is characterized by specifying.

According to the seventeenth aspect, in the document image,
Determines whether each character area is a Japanese area or an English area.
A Japanese-English determination method for a document image to be
A circumscribed rectangle is generated from an image obtained by reducing the
Rectangles that are in a positional relationship are merged, and the aspect ratio of the merged rectangles
Each character area is a Japanese area based on the histogram of
Or English language domain
You.

According to the eighteenth aspect of the present invention, the document image
Determines whether each character area is a Japanese area or an English area.
A Japanese-English determination method for a document image to be
A circumscribed rectangle is generated from an image obtained by reducing the
Rectangles that are in a positional relationship are merged, and the aspect ratio of the merged rectangles
Each character area is a Japanese area based on the histogram of
Or in the English area,
For unknown areas that cannot be
Calculate the characteristic value of Japanese and the characteristic value of English
Calculating which characteristic value the unknown area is closer to
To determine whether the area is Japanese or English
It is characterized by that.

According to the nineteenth aspect, it is determined whether the document image is a Japanese document image or an English document image.
It is characterized in that a document recognition process is performed according to the determination result.

In the twentieth aspect, a document image is divided into a plurality of character areas, and it is determined whether each of the divided character areas is a Japanese document area or an English document area. It is characterized in that a document recognition process is performed in accordance with.

According to the twenty-first aspect, a plurality of determination methods are used to determine whether each character area in a document image is a Japanese area or an English area. It is a computer-readable recording medium that records a program for causing a computer to realize a function of determining whether an area is an area and a function of obtaining a final determination result by comparing the plurality of determination results. .

In the invention according to claim 22, in order to determine whether the document image of each character area or each page in the document image is a Japanese area or an English area, the document image is reduced. A function of classifying the connected component based on the length of the black pixel connected component in the generated character area or page, and each of the character areas or pages in the Japanese area based on the total value of the classification result. It is a computer-readable recording medium that records a program for causing a computer to realize a function of determining whether a region is an English language region.

According to the twenty-third aspect of the invention, in order to determine whether each character area in the document image is a Japanese area or an English area, or to determine whether each page is composed of a plurality of character areas, In order to determine whether the document image is a Japanese document image or an English document image, a function of detecting a line from the character area and a block that is extracted by integrating a circumscribed rectangle close to the line from the character region A function for determining whether each block is a Japanese area, an English area, or a non-determinable area; a function for totalizing the determination result for each block or for each page; Based on the value, a computer implements a function of determining whether each of the character areas is a Japanese area or an English area or a function of determining whether each page is a Japanese or English document image. Let It is characterized by a computer-readable recording medium recording a program for.

According to the twenty-fourth aspect of the present invention, a function of determining whether a document image is a Japanese document image or an English document image or a function of dividing a document image into a plurality of character areas, A computer-readable recording medium that records a program for causing a computer to implement a function of determining whether a document area is a Japanese document area or an English document area and a function of performing document recognition processing according to the determination result It is characterized by having.

According to the twenty-fifth aspect, in the document image,
A function to cut out lines from each character area
Histogram of the percentage of each rectangle height in a row or rectangle in a row
Hist of the ratio of the height of each rectangle in the row to the maximum height of the shape
Whether each character area is a Japanese area or English based on the gram
A function to determine whether an area is an area
For unknown areas that cannot be
A function to calculate Japanese characteristic values and English characteristic values,
Calculating which characteristic value the unknown region is close to
To determine whether the area is Japanese or English
To make a computer realize the function of
Is a computer-readable recording medium that records
It is characterized by:

According to the twenty-sixth aspect, the document image
A function to generate a reduced image, and a character area from the reduced image.
Function for generating a circumscribed rectangle from the character area
And a function to integrate rectangles in a predetermined positional relationship
And the above sentence based on the histogram of the aspect ratio of the rectangle after integration.
Determine if the character area is Japanese or English
A program to realize the functions
Be a computer-readable recording medium
It is characterized by.

[0034]

DESCRIPTION OF THE PREFERRED EMBODIMENTS One embodiment of the present invention will be specifically described below with reference to the drawings. <Embodiment 1> FIG. 1 shows the structure of Embodiment 1 of the present invention.
In the figure, 101 is an image input unit for inputting a document image, 102 is an image reduction unit for reducing an input document image,
103 is a connected component extracting means for extracting connected components from the document image; 104 is a region generating means for generating a character region by classifying and integrating the extracted connected components;
5 is a Japanese-English discriminating means for discriminating between Japanese and English in a character area unit or a page unit, 106 is a control unit for controlling the whole, and 107 is various kinds of input document image data, connected component data, area data, etc. A data storage unit 108 for storing data is a data communication path, and 109 is a data communication unit connected to a host or the like via a network, a line, or the like.

FIG. 2 is a flowchart showing the entire process according to the first embodiment of the present invention. Hereinafter, the processing operation of the present invention will be described with reference to FIG. First, the image input unit 101
Obtains a document image by reading the document (step 201). This image input means is, for example, a scanner,
It may be a facsimile or the like, and an image may be obtained from another device via a network via the data communication means 109.

Next, the image reducing means 102 reduces the input document image (step 202). This process
For example, a process of OR-reducing the input document image to about 1/8. That is, the process is to reduce 8 × 8 pixels to one pixel, and if there is even one black pixel in 64 pixels, the reduced pixel is set to a black pixel.

The connected component extracting means 103 extracts a black pixel connected component from the reduced image (Step 203). The region generating means 104 classifies the extracted connected components and integrates them to generate a character region (step 204). As this area generation method, for example, a known method described in JP-A-6-20092 may be used. At this time, information on the connected components constituting each character area is stored and held in the data storage unit 107.

Subsequently, for the generated character area, the Japanese / English determining means 105 determines whether the character area is Japanese or English (step 205).

In step 202, the adjacent black pixels are merged by OR-reducing the image. Here, English sentences have the feature that there is a space between words and the space between characters in a word is very narrow. On the other hand, in Japanese, the character spacing does not change significantly except before and after punctuation.

FIG. 3 shows image examples of English and Japanese sentences and their circumscribed rectangles. The result obtained by reducing the English image 301 and extracting the connected components as a circumscribed rectangle is the circumscribed rectangle 30.
2 (note that the circumscribed rectangle 302
Should be smaller than the image 301, but are represented in the same size here). In an English image, a connected component is formed by fusing each word.

For examples of Japanese images 303 and 305,
Similarly, when the connected component is extracted by being reduced, and is expressed by a circumscribed rectangle, the circumscribed rectangles 304 and 306 are obtained, respectively.

In the case of English sentences, since the number of characters constituting a word is constant to some extent, there is a feature that the circumscribed rectangle having an aspect ratio of about 2 to 6, or 7 is increased. On the other hand, in the case of Japanese, there is a feature that a long rectangle which is hard to appear in English sentence occurs as shown by a circumscribed rectangle 304 and, on the contrary, many small rectangles such as a circumscribed rectangle 306 occur.

Therefore, the above-mentioned connected component rectangle is called “short”,
It is classified into three types of "medium" and "long", and these are totaled for each character area. FIG. 4 is a processing flowchart of the Japanese-English determination according to the first embodiment. The process of FIG. 4 is performed for each character area. When the row direction is horizontal, for example, the classification of rectangles is “short” when the width / height is 2 or less, “medium” when the width / height is 2 to 6, and “long” when the width / height is more than 2 (step). 401).
Then, the classification results in the character area are totaled (step 402), and it is determined whether the character area is Japanese or English (step 403). Here, the number of “short” rectangles is SC
NT, the number of “medium” rectangles is NCNT, and the number of “long” rectangles is L
Assuming CNT, the determination of Japanese or English is performed as shown in FIG. 8 (detailed flowchart of step 403).

First, LCNT / (NCNT + SCNT)
> Thl is checked (step 80)
1). Th1 is a predetermined threshold value.
It should be about 3. If this conditional expression holds, it means that there are many long rectangles, and it is determined that the character area is a Japanese area (step 804).

Next, when No is determined in step 801, it is checked whether NCNT / (LCNT + SCNT) <Th2 is satisfied (step 802). Th2 is also a predetermined threshold value, for example, 3. If this conditional expression holds, it means that there are few middle rectangles, and it is determined that the character area is a Japanese area (step 8).
04). If neither condition is satisfied, it is determined that the region is an English region (step 803).

<Embodiment 2> In the above-described embodiment 1, the judgment of Japanese or English is made for each character area. In this case, the number of characters may be very small depending on the character area. In such a case, the number of rectangles cannot be obtained sufficiently, so that it may be difficult to perform Japanese-English determination at the ratio of the number of rectangles.
The second embodiment is an embodiment that considers a case where the number of rectangles is not sufficient.

FIG. 5 shows a processing flowchart of the second embodiment. The Japanese-English discriminating means 105 checks whether or not the number of rectangles in the totaled area is sufficient (that is, whether or not the number is equal to or larger than a predetermined threshold Th) (step 501). Japanese-English discrimination using OCR described in JP-A-6-150061 is performed (step 5).
03). In this case, since the number of characters is small, even if OCR processing is performed, the increase in processing time is small. If the number of rectangles is sufficient, Japanese and English are identified by the rectangle length described in the first embodiment (step 502).

<Embodiment 3> Next, an embodiment 3 for performing Japanese-English discrimination on a page basis will be described. 6 and 7 show a detailed flowchart of step 205 according to the third embodiment.
The method shown in FIG. 6 counts the number of “short”, “medium”, and “long” rectangles not for each character area but for the entire page (steps 601 and 602), and uses the result to store the date in page units. An English determination is made (step 603). This Japanese / English determination method is performed according to the processing flowchart of FIG. The thresholds Th1 and Th2 at this time may be different from those in the case of processing in units of character areas.

In the method shown in FIG. 7, Japanese or English is determined for each character area (step 702), and based on the result, Japanese or English is determined for the page (step 703). Specifically, the number of regions determined to be Japanese regions is Jn, and the number of regions determined to be English regions is En, where Jn> En is a Japanese page, and En> Jn is an English page. J
If n = En, rejection may be performed, or the determination may be made in either Japanese or English.

<Embodiment 4> A Japanese-English identification method using a feature different from that of the above-described embodiment will be described. FIG.
4 shows a configuration of a fourth embodiment. The difference from the first embodiment is that a line segmentation unit 902, a block extraction unit 903, and an in-block character type discrimination unit 904 are provided. Other components are the same as those of the first embodiment. FIG.
3 shows a processing flowchart.

First, the line cutout unit 902 cuts out a line from the character area of the document image (step 1001,
1002). Japanese Patent Application Laid-Open No. 6-2009
In the case of using the technology described in Japanese Patent Publication No. 2 (1993), line information is obtained at the stage of extracting a region, and this may be used. Segmentation of Document Image Using ”(Akiyama et al.,
August 1986, Vol. J69-D No. The method using projection described in 8) may be used.

Next, the block extracting unit 903 extracts a block corresponding to a word (step 1003). As the block extraction method, the present applicant has previously disclosed in Japanese Patent Application No. 8-347.
The method proposed in No. 81 may be used. That is, the block extracting unit 111 detects a circumscribed rectangle in the row data, and combines the circumscribed rectangle into block data. The method of combining the block data is as follows. A histogram of the character rectangle intervals (one rectangle is not yet determined to be one character. Therefore, in the case of kanji, a rectangle separated from bias and structure often becomes one rectangle) is obtained. FIG. 18 shows the extracted character rectangles and the distance between the rectangles. FIG. 19 shows a histogram of rectangular intervals.

In this histogram, the shortest distance peak tends to appear in the interval between the bias and formation of kanji and the distance between characters in the same word of proportional alphabetic characters. Even if these are integrated, different character types rarely enter a block, so that block data is formed by integrating them. By performing this processing, proportional words and kanji characters in which one character is separated (that is, composed of bias and structure) are integrated into one.

The peak with the longest distance often appears at the distance between words and the distance between punctuation marks and the next character.
These are often used at the boundary when the character type changes (especially the distance between words), and it is desired to avoid the same block. Therefore, processing is performed so that character rectangles having a distance equal to or longer than the longest peak value are not placed in the same block.

Further, the distance (A, B) between the target rectangle and the adjacent rectangles is measured. When the difference (A−B) is equal to or larger than a predetermined threshold, the rectangles having the longer distance are not integrated. Process to combine rectangles with shorter distances. FIG. 20 is a diagram illustrating a case where rectangles are not integrated at a position where the difference between the rectangles is large. In FIG. 20, three blocks are formed because rectangles are not integrated at a position where the difference is larger than a predetermined threshold. By performing such processing, even in a proportional English sentence or the like, even if the distance between words is absolutely short, the distance between characters must be different from the distance between characters, so that only one word can be integrated together. Even in a proportional font, Japanese kanji portions are arranged at relatively equal intervals, so that it is convenient to combine Japanese sentences.

By using the above-described block extraction method, in the case of an English sentence, unlike a Japanese document, words are separated by a space equivalent to a half-width, so that block data is mixed with other character types. That can be avoided.

Subsequently, a character type discriminating unit 904 in a block is performed.
Performs a Japanese-English determination for each block (step 100).
4). This may also use the method of the above-mentioned application. In other words, the intra-block character type determination unit 904 determines whether the block grouped by the above process is a Japanese character or an alphanumeric character. The inside of the block is determined as the same character type. This character type is determined as follows. That is, when the number of black runs in the vertical direction or the number of black-and-white inversions of the rectangle in the block is equal to or greater than a predetermined threshold, the rectangle is identified as a Japanese character, and the rectangle in the extracted block in the vertical direction is Identify alphabetic characters based on coordinate values.
FIGS. 21A and 21B show specific examples of the number of vertical runs for Japanese and English characters. In an ideal case where there is no noise in alphanumeric characters, four runs can be made at maximum with the letter “g” (FIG. 21B). Therefore, when five or more runs are counted, the language is set to Japanese. In the case of the character "image" shown in FIG. 21A, the number of runs in the vertical direction changes as indicated by the number below the character.

The English / Japanese discriminating means 905 sums up the discrimination results for each block and performs Japanese / English discrimination of the area (step 1).
005). Here, the number of blocks determined as Japanese is JCNT, the number of blocks determined as English is ECNT,
The number of blocks determined to be undefined is defined as NCNT. FIG.
1 is a detailed flowchart of step 1005. If JCNT * Th3> ENCT, it is determined that the language is Japanese (steps 1101 and 1105).
When CNT> JCNT, it is determined to be English (1102,
1104). Otherwise, it is rejected (step 1103). The threshold value Th3 is, for example, 2.

<Embodiment 5> In Embodiment 4 described above, the judgment of Japanese or English is made in units of character areas. In this case, the number of characters may be very small depending on the character area. In such a case, the number of rectangles cannot be obtained sufficiently, so that it may be difficult to perform Japanese-English determination based on the ratio of the number of block determination results. Embodiment 5 is an embodiment in which the number of blocks is not sufficient.

FIG. 12 shows a processing flowchart of the fifth embodiment. The Japanese-English discriminating means 105 checks whether or not the total number of blocks in the character area is sufficient (that is, whether or not the number is equal to or greater than a predetermined threshold Th) (step 1201). Japanese-English distinction using OCR described in Japanese Patent Application Laid-Open No. Hei 6-150061 is performed (step 1203). In this case, since the number of characters is small, even if OCR processing is performed, the increase in processing time is small. Then, if the number of blocks is sufficient, Japanese and English are identified based on the determination result for each block described in the fourth embodiment (step 1202).

<Embodiment 6> In Embodiment 6, the Japanese / English distinction for each character area in Embodiment 4 is changed to Japanese / English distinction in page units. FIG. 6 is a processing flowchart of the sixth embodiment.
7 is used.

In the processing of FIG. 6, JCNT, ECN
The total of T and NCNT is calculated not for each character area but for the entire page, and the result is used to determine Japanese or English by the processing method of FIG. 11 described above. At this time, Th3 may be different from the case of the character area unit.

In the processing shown in FIG. 7, first, a determination is made for each character area, and the result is used to determine whether the page is Japanese or English. Specifically, the number of areas determined to be Japanese
n, the number of areas determined to be English areas as En, and Jn
If> En, it is determined to be a Japanese page, and if En> Jn, it is determined to be an English page. When Jn = En, rejection may be used, or any of Japanese and English may be used.

<Embodiment 7> In the embodiment 7, when performing the Japanese / English distinction for each character area or for each page, as shown in FIG.
1) and Japanese-English determination processing (step 1302) using the determination result for each block, to determine Japanese and English, respectively. Then, from the respective determination results, a determination is finally made between Japanese and English (step 1303).

If both are determined to be Japanese or English, the final result may be determined to be Japanese or English as it is. If any are determined to be rejected,
The result of the determination that is not reject is the final result.

If the result of the determination is that one is Japanese and the other is English, and the results do not match, one of the following is determined. (1) Reject. (2) The two confidence factors are calculated, and the result with the larger value is adopted. As the certainty factor of the discrimination method using the rectangular length, for example, LCNT / (NCNT + SCNT)> Thl, Thl
LCNT / (NCNT + SCNT) when = 0.3
* Value of 2.5 (upper limit is 1) NCNT / (LCNT + SCNT) <Th2, Th2
If = 3, (LCNT + SCNT) / NCNT *
NCNT / (LCNT + SCNT)> Th2 when the value of 2.5 (the upper limit is 1)
If N = 3, NCNT / (LCNT + SCNT) *
0.33 (upper limit is 1).

As a certainty factor of the discrimination method using the discrimination result for each block, for example, when JCNT * Th3> ECNT and Th3 = 2, the value of JCTN / (ECNT * 3) (the upper limit is 1)
If ECNT> JCNT, ECNT / JCNT *
0.7 (however, the upper limit is 1).

<Eighth Embodiment> FIG. 14 shows the structure of the eighth embodiment. FIG. 15 shows a processing flowchart of the eighth embodiment. In this embodiment, for the entire page of the input document, the Japanese-English discriminating unit 1412 uses the third embodiment described above.
Using the method of No. 6, Japanese-English discrimination processing is performed to determine whether the page is in Japanese or English (steps 1501 and 15).
02) Based on the determination result, the selection unit 1403 selects the English document recognition unit 1404 or the Japanese document recognition unit 1405, performs the document recognition processing of the selected language (steps 1504 and 1505), and outputs the recognition result. The data is output to an output unit such as a display (step 1506).

Since the attributes of Japanese and English are different, it may be better to switch the area division processing and the font identification processing. Therefore, the document recognition unit of this embodiment includes not only the character recognition processing but also the above-described area division processing and font identification processing.

<Embodiment 9> FIG. 16 shows the structure of the ninth embodiment, and FIG. 17 shows a processing flowchart of the ninth embodiment. The difference from the eighth embodiment is that Japanese-English identification is performed for each character area. For this purpose, the area dividing unit 1602 divides the input document into character areas (steps 1701 and 170).
2). Here, the region dividing unit uses a region dividing method applicable to both Japanese and English. After the division process, the Japanese / English discriminating unit 1603 performs a Japanese / English identification process for each character area using, for example, the method of the first embodiment described above (step 170).
4) Based on the discrimination result, the selecting unit 1604 selects the English document recognizing unit 1605 or the Japanese document recognizing unit 1606, performs a document recognizing process of the selected language (steps 1705 and 1706), and outputs the recognizing result. The output is output to an output unit 1607 such as a display (step 1707).
The document recognition unit according to the ninth embodiment performs a font identification process in addition to the document recognition process.

<Embodiment 10> In each of the embodiments described above, Japanese and English are determined using black pixel connected components and rectangular lengths as feature amounts. However, the determination method using the black pixel connected component takes a long processing time, and the method using the rectangular length may cause high rejection. There is also a method of distinguishing between Japanese and English based on the peak position of the frequency distribution of the relative position in the upper and lower lines of the circumscribed rectangle (Japanese Patent Publication No.
However, when a document having an inclination is input, there is a problem that the frequency distribution greatly changes and the identification accuracy is reduced.

Therefore, in this embodiment, the line height is
By distinguishing between Japanese and English using the histogram of the height of the circumscribed rectangle in the line, Japanese and English are accurately and quickly identified for each area of the document image. Then, for an area that cannot be determined by the above-described Japanese-English identification method, Japanese-English identification is performed using another method.

FIG. 22 shows the structure of the tenth embodiment. FIG. 23 is an overall processing flowchart of the tenth embodiment. First, the image input unit 2201 obtains a document image by reading a document (step 2301).
The image input unit is, for example, a scanner, a facsimile, or the like, and an image may be obtained from another device via the data communication unit 2207 via a network.

Next, the area generating means 2202 generates a character area (step 2302). For example, a method described in JP-A-6-20092 may be used as the region generation method. Next, the line segmentation means 2203
Extracts a line for character recognition from a character area. That is, the circumscribed rectangles of the characters are obtained, and they are integrated to generate a line (step 2303). Japanese-English identification means 22
04 performs Japanese-English identification for the generated character area (step 2304).

The distinction between Japanese and English is performed as follows. FIG.
FIG. 7 is a detailed flowchart of the Japanese-English identification (step 2304). FIG. 24 shows an example of a cut-out line and a circumscribed rectangle in the line. First, the frequency distribution of the ratio of the height of the circumscribed rectangle in the row to the row height is calculated (step 270).
1, 2702). The line height is lineheight, and the rectangle height is height. Weight ratio
Let e = height * 100 / lineheight. In the case of a document having a slope as shown in FIG. 25, the maximum value of the height of the rectangle of the line may be used as lineheight instead of the line height in order to more accurately identify Japanese and English. That is, with respect to an input document having a slope, Japanese and English are identified based on a histogram of the ratio of the height of each circumscribed rectangle in the line to the maximum height of the rectangle in the line.

If the above-mentioned ratio heightrate is, for example, 80 or more, the number of rectangles is set to lcnt, and
The number of rectangles when the rate is, for example, 70 or more and less than 80 is ncnt, and the number of rectangles when the heightrate is, for example, 40 or more and less than 70 is scnt. Lcnt, ncnt, scnt for all rectangles in the character area
Ask for.

FIG. 26 shows an example of the number of rectangles checked for a Japanese document and an English document. In general, Japanese is lcnt
English tends to have a large scnt.
Therefore, predetermined thresholds thJ and thE are set, and lcnt
If / scnt> thJ, it is determined that the language is Japanese (step 2
703), if lcnt / scnt <thE, it is determined that the language is English (step 2704). Otherwise, it is set as an unknown area (step 2705).

The above-mentioned unknown area can be distinguished between Japanese and English by using a statistical method. FIG. 28 is a detailed processing flowchart for an unknown area. For example, feature values lcnt, nc of the Japanese region and the English region in advance
nt and scnt are normalized, and the mean and the inverse matrix of the covariance matrix are obtained for Japanese and English, respectively. Then, the Mahalanobis distance is calculated for each of Japanese and English using the average value and the inverse matrix of the covariance matrix (steps 2801 and 2802).

When the Japanese Mahalanobis distance is Dj and the English Mahalanobis distance is De, the predetermined threshold is M
Assuming e and Mj, English is determined when Dj / De> Me (step 2803), and Japanese is determined when Dj / De <Mj (step 2804). If none of the conditions is satisfied, it is determined that the area is unknown (step 280).
5). Instead of the Mahalanobis distance described above, a Euclidean distance with an average value or a city block distance may be used.

Further, for an area determined to be unknown, Japanese-English discrimination is performed using the certainty factor of English sentence recognition. FIG. 29 is a detailed processing flowchart of step 2805. The confidence is calculated by English sentence recognition (step 2901). Next, for the calculated certainty, for example, the number of words having a certainty of 60% or more is Good, the number of words having a certainty of less than 60% is 0, and the number of words having a certainty of 0 is Zero. (Step 2902).

Assuming that the judgment value for Japanese-English discrimination is Value, Value = Good / (Good + Bad + Ze
lo) (step 2903), and when the value exceeds a predetermined threshold theeocr (step 2904),
It is determined to be English, and if it is less than that, it is determined to be Japanese.

The weight may be assigned to Zero. Z
If ero is, for example, three Bads, Value
Since Bad = Bad + Zero × 3, Value = Good / (Good + Bad), and if the value exceeds the threshold theeocr, it can be determined that the language is English, and if it is less than that, the language can be determined to be Japanese. In this way, even in an area where the number of characters for Japanese / English identification is small, the English / Japanese identification is performed with the certainty based on the English sentence recognition.

<Embodiment 11> In this embodiment, a circumscribed rectangle is generated from an image obtained by reducing an input document image, an appropriate integration is performed between the generated rectangles, and a histogram of the rectangular length aspect ratio after the integration is obtained. This is an embodiment in which Japanese-English discrimination is performed with higher accuracy by using this method.

FIG. 30 shows the structure of the eleventh embodiment. FIG. 31 is an overall processing flowchart of the eleventh embodiment. Image input means 30 in the same manner as in the above-described embodiment.
01 is input to the image reducing unit 30.
02 (steps 3101 and 310
2). In this processing, for example, the document image is ORed to about 1/4.
Compression (4 × 4 pixels are reduced to 1 pixel, and if there is at least one black pixel in 16 pixels, the reduced image is black).

Next, the area generating means 3003 generates a character area (step 3103). For example, a method described in JP-A-6-20092 may be used as the region generation method. Subsequently, the rectangle integration means 3004
Performs integration of rectangles so that the characteristics of Japanese and English can be well expressed (step 3104). For example, as shown in FIG. 32, when the vertical coordinates of the y-coordinates (vertical direction) of the rectangles 1 and 2 are close and the x-coordinates of the adjacent rectangles 1 and 2 are very close (for example, the horizontal distance between the rectangles). Is smaller than the distance corresponding to the English space), the rectangle is integrated. Also, for example,
As shown in FIG. 33, the left rectangle 1 is replaced with the right rectangle 2 by y.
The rectangles 1, which are in a positional relationship encompassed by coordinates and are adjacent to each other,
2 is very close (eg, the horizontal distance between rectangles is less than the distance corresponding to English space)
Merge rectangles.

Then, using the rectangular aspect ratio (rectangular long vertical / rectangular long horizontal), four rectangles of a long rectangle, a medium rectangle, a small rectangle, and a very small rectangle are obtained.
It is divided into two features (FIG. 34). In general, Japanese has a high proportion of long rectangles, and English has a high proportion of medium rectangles. Utilizing this difference in characteristics, the English-Japanese identification means 3005 creates an identification determination formula and performs Japanese-English identification (step 3105). FIG. 35 is a detailed flowchart of the Japanese-English identification processing.

For example, the number of long rectangular areas in the area lcnt The number of medium rectangular areas in the ncnt area The number of small rectangular areas in the ncnt area The number of small rectangular areas in the scnt area The number of small rectangular areas sscnt (in the case of noise, Is calculated (step 3501), and the ratio of the long rectangle in the area ratio1 = lcnt / (ncnt + sc)
nt) is calculated (step 3502), and the ratio of the middle rectangle in the area ratio2 = ncnt / (lcnt + scn)
t) is calculated (step 3503). When calculating the above ratio, sscnt was ignored as noise.

Then, ratio is defined as x coordinate, ratio
Let o2 be the y coordinate, erroneous identification is reduced as much as possible, and the overlapped portion is divided into a Japanese region, an English region, and a reject region so as to be rejected. For example, ratio2
If / ratio> thE, it is determined that the region is an English region (step 3504), and ratio2 / ratio <thJ
If so, it is determined that the area is a Japanese area (step 3505), and the other areas are unknown in Japanese and English (step 3506).
Here, thE and thJ are predetermined thresholds.

In the same manner as in the tenth embodiment, the area determined to be Japanese / English unknown is identified using a statistical method. For example, the feature value lcn of the Japanese region and the English region in advance
The t, ncnt, and scnt are normalized, and the average value and the inverse matrix of the covariance matrix are obtained in Japanese and English, respectively. The Mahalanobis distance for each of Japanese and English is calculated using the mean and the inverse matrix of the covariance matrix. When the Mahalanobis distance in Japanese is Dj and the Mahalanobis distance in English is De, and given thresholds are Me and Mj, Dj / De> M
If e, English is determined, and if Dj / De <Mj, Japanese is determined. If none of the conditions is satisfied, it is determined to be unknown. Note that, instead of the Mahalanobis distance, a Euclidean distance with an average value or a city block distance may be used.

<Embodiment 12> The present invention is not limited to the above-described embodiment, and can be realized by software. When the present invention is implemented by software, as shown in FIG. 36, a CPU, a memory, a display device,
Hard disk, keyboard, CD-ROM drive,
Prepare a computer system consisting of a scanner, etc.
On a computer-readable recording medium such as a CD-ROM, a program for realizing the Japanese-English determination function and the document recognition function of the present invention is recorded. A document image or the like input from an image input unit such as a scanner is temporarily stored in a hard disk or the like. Then, when the program is started, the temporarily stored document image data is read, a Japanese-English determination process and a document recognition process are executed, and the results are output to a display or the like.

[0091]

As described above, according to the present invention, a plurality of determination methods are used in combination, so that Japanese and English can be determined with high accuracy. Further, it is possible to accurately determine Japanese and English for each character region in the document image, and it is possible to accurately determine Japanese and English for each page of the document image. Furthermore, since an appropriate document recognition process is performed on a document image determined to be Japanese or English, a highly accurate recognition result can be obtained.

[Brief description of the drawings]

FIG. 1 shows a configuration of a first exemplary embodiment of the present invention.

FIG. 2 shows an overall processing flowchart of Embodiment 1 of the present invention.

FIG. 3 shows image examples of English sentences and Japanese sentences and their circumscribed rectangles.

FIG. 4 is a processing flowchart of Japanese-English determination according to the first embodiment.

FIG. 5 shows a processing flowchart of a second embodiment.

FIG. 6 shows a first detailed flowchart of step 205 according to the third embodiment.

FIG. 7 shows a second detailed flowchart of step 205 according to the third embodiment.

FIG. 8 shows a detailed flowchart of step 403.

FIG. 9 shows a configuration of a fourth embodiment.

FIG. 10 shows a processing flowchart of a fourth embodiment.

FIG. 11 is a detailed flowchart of step 1005.

FIG. 12 shows a processing flowchart of a fifth embodiment.

FIG. 13 shows a processing flowchart of a seventh embodiment.

FIG. 14 shows a configuration of an eighth embodiment.

FIG. 15 shows a processing flowchart of the eighth embodiment.

FIG. 16 shows a configuration of a ninth embodiment.

FIG. 17 shows a processing flowchart of the ninth embodiment.

FIG. 18 shows an extracted character rectangle and a distance between the rectangles.

FIG. 19 shows a histogram of rectangular intervals.

FIG. 20 is a diagram illustrating a case where rectangles are not integrated at a position where the difference between the rectangles is large.

FIGS. 21A and 21B show specific examples of the number of vertical runs for Japanese and English characters.

FIG. 22 shows a configuration of a tenth embodiment.

FIG. 23 is an overall processing flowchart of the tenth embodiment.

FIG. 24 shows an example of a cut-out line and a circumscribed rectangle in the line.

FIG. 25 shows an example of a line when a document is inclined and a circumscribed rectangle in the line.

FIG. 26 shows an example of the number of rectangles examined for a Japanese document and an English document.

FIG. 27 is a detailed processing flowchart of Japanese-English identification (step 2304).

FIG. 28 is a detailed processing flowchart for an unknown area.

FIG. 29 is a detailed processing flowchart of step 2805.

FIG. 30 shows a configuration of an eleventh embodiment.

FIG. 31 is an overall processing flowchart of an eleventh embodiment.

FIG. 32 shows an example of integrating rectangles.

FIG. 33 shows another example of merging rectangles.

FIG. 34 shows rectangles classified into four types.

FIG. 35 is a detailed processing flowchart of Japanese-English identification processing in Example 11;

FIG. 36 shows a structure of a twelfth embodiment.

DESCRIPTION OF SYMBOLS 101 Image input means 102 Image reduction means 103 Connected component extraction means 104 Area generation means 105 Japanese / English discrimination means 106 Control unit 107 Data storage unit 108 Data communication path 109 Data communication means

Claims (15)

[Claims] 1. A method for determining whether a character region in a document image is a Japanese region or an English region, the method comprising the steps of: A method for determining Japanese or English for a document image, which determines whether a document image is in an English area or not, and obtains a final determination result by comparing the plurality of determination results. 2. A method for judging whether or not each character area in a document image is a Japanese area or an English area, which is generated by reducing the size of the document image. Classifying the connected component based on the length of the black pixel connected component in the character region, and determining whether each of the character regions is a Japanese region or an English region based on a total value of the classification result. Japanese-English determination method for document images characterized by 3. The Japanese-English determination of a document image according to claim 2, wherein a different determination method is used when the number of connected black pixels in the generated character area does not satisfy a predetermined condition. Method. 4. A method for determining whether a document image of each page is a Japanese document image or an English document image in Japanese or English, which is generated by reducing the document image. Classifying the connected component based on the length of the black pixel connected component in the page, and determining whether each page is a Japanese region or an English region based on a total value of the classification result. Japanese and English judgment method of the document image to be used. 5. A method according to claim 1, wherein the page is composed of a plurality of character areas, and the document image of each page is a Japanese document image or an English document image. Classifying the connected component based on the length of the black pixel connected component in the character region generated by reducing the image, and determining whether each of the character regions is a Japanese region based on the total value of the classification result; A method for determining whether a page is a Japanese area or an English area based on a determination result of whether the page is an English area or not, based on the determination result. 6. A method for determining whether a character region in a document image is a Japanese region or an English region, the method comprising the steps of: detecting a line in the character region;
A block is extracted by integrating adjacent circumscribing rectangles from the line, and it is determined whether each block is a Japanese area, an English area, or an undeterminable area, and the determination result is determined by the block. A method for determining whether each character area is a Japanese area or an English area based on the total value. 7. The method according to claim 6, wherein a different judgment method is used when the number of extracted blocks does not satisfy a predetermined condition. 8. A method for determining whether a document image of a page is a Japanese document image or an English document image, wherein the document image includes a plurality of character areas, Detects a line from the area, integrates adjacent circumscribed rectangles from the line, extracts blocks, and determines whether each block is a Japanese area, an English area, or an undeterminable area And summing up the determination results on a page-by-page basis, and determining whether each of the pages is a Japanese document image or an English document image based on the total value. Method. 9. A method for determining whether a document image of a page is composed of a plurality of character areas and determining whether the document image of each page is a Japanese document image or an English document image. Detects a line from the area, integrates adjacent circumscribed rectangles from the line, extracts blocks, and determines whether each block is a Japanese area, an English area, or an undeterminable area Then, the determination result is totaled for each character region, and it is determined whether the region is a Japanese region or an English region for each character region based on the total value, and the determination result is totaled for each page. A method for determining whether each page is a Japanese document image or an English document image based on a value. 10. A document recognition method comprising: determining whether a document image is a Japanese document image or an English document image; and performing a document recognition process according to a result of the determination. 11. A document image is divided into a plurality of character areas,
A document recognition method comprising: determining whether each of the divided character regions is a Japanese document region or an English document region, and performing a document recognition process according to the determination result. 12. A plurality of determination methods are used to determine whether each character area in a document image is a Japanese area or an English area using a plurality of determination methods. And a computer-readable recording medium storing a program for causing a computer to realize a function of performing a final determination result by comparing the plurality of determination results. 13. A character region generated by reducing the size of a document image in order to determine whether the character image in each character region or each page in the document image is a Japanese region or an English region. Alternatively, a function of classifying the connected component based on the length of the black pixel connected component in the page, and each character region or each page is a Japanese region or an English region based on a total value of the classification result. A computer-readable recording medium on which a program for causing a computer to realize the function of determining whether the program has been stored is recorded. 14. A method for determining whether each character area in a document image is a Japanese area or an English area, or
A page is composed of a plurality of character areas, and a function of detecting a line from the character area to determine whether the document image of each page is a Japanese document image or an English document image,
A function of extracting blocks by integrating adjacent circumscribed rectangles from the row, a function of determining whether each block is a Japanese area, an English area, or an undeterminable area; A function for summing up the results for each block or for each page, and a function for judging whether each of the character areas is a Japanese area or an English area based on the total value, or each page is a Japanese document image. A computer-readable recording medium that records a program for causing a computer to realize a function of determining whether a document image is an English document image. 15. A function for determining whether a document image is a Japanese document image or an English document image, or dividing a document image into a plurality of character regions, and dividing each of the divided character regions by a Japanese document region. A computer-readable storage medium storing a program for causing a computer to realize a function of determining whether a document area is a document area or not and a function of performing document recognition processing according to the determination result.