JP2000148788A - Title region extraction device and title region extraction method from document image, and document retrieval method - Google Patents

Abstract

(57) [Summary] [PROBLEM] To automatically extract a character string region having a large number of points as a title by using a characteristic unique to a title as a point without depending on a specific document format, and to accurately extract the title and to obtain a document. Improve convenience when searching. SOLUTION: A character recognizing unit 102 for recognizing a character string in a character string rectangle cut out by an area identification unit 101, and for each character in the character string rectangle for the character string rectangle. Font identification unit 103 for identifying a font
And a natural language analysis unit 104 for analyzing the likelihood of a natural language title based on the character code obtained from the result of recognition by the character recognition unit 102, and for the character string rectangle, centering, underlining, the size of the character rectangle, etc. And a point assigning section 105 for assigning a point to the title likeness.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for accurately expressing document contents from a database of document image data input from an image input device such as a facsimile or an image scanner in order to improve the convenience of retrieval. The present invention relates to an apparatus and method for extracting a title area from a document image that extracts an area in a document as a title area, and a document search method.

[0002]

2. Description of the Related Art Conventionally, when retrieving a document image, an operator manually operates the content of the document separately from the input of the document image from an image input device in order to improve the convenience of later retrieval. Is extracted / created and added to the title information and keyword information that accurately expresses, and a fixed position (character string) in the document is cut out as a title / keyword for a fixed-form document.

[0003] Further, a reference technical document for extracting a title from an irregular document using only layout features is disclosed in, for example, "Apparatus and method for extracting title from document image" in JP-A-9-134406. Kaihei 5-2744
No. 71, "Title area extraction processing method of image document" is disclosed.

[0004]

However, in the prior art as described above, the addition of title information and keyword information by the operator increases the amount of work as the amount of documents increases. This will increase the burden. In addition, when automatic extraction at a specific position is performed, only fixed-form documents are targeted, so that it cannot be used for non-standard-size documents, and there is a problem that convenience is lacking.

[0005] Japanese Unexamined Patent Application Publication No. 9-134 disclosed in the prior art
In Japanese Patent Publication No. 406 and Japanese Patent Laid-Open Publication No. Hei 5-274471, titles are extracted by paying attention only to layout features, and therefore, the hit ratio of titles that accurately express the contents of a document cannot always be satisfied. However, there were problems such as hindrance to later document retrieval.

The present invention has been made in view of the above, and uses a characteristic unique to a title as a point without depending on a specific document format, thereby automatically extracting a character string area having a large number of points as a title. The purpose of the present invention is to improve the accuracy of title extraction and the convenience at the time of document search.

[0007]

According to a first aspect of the present invention, there is provided an apparatus for extracting a title area from a document image, comprising the steps of extracting a character string area from a document image input from an image input apparatus. It has an area identification unit that cuts out a rectangle,
In a title region extracting apparatus for performing a title-likeness point calculation based on the attribute of the character string region and extracting a title from a document image, a character string rectangle extracted by the region identifying means is compared with the character string rectangle. Character recognizing means for recognizing characters in the character string; font identifying means for recognizing a font of each character in the character string rectangle cut out by the area identifying means; and recognition of the character recognizing means. Natural language analysis means for analyzing the likelihood of a natural language title based on the character code obtained as a result; and a character string rectangle cut out by the area identification means using centering, underlining, the size of the character rectangle, and the like. Pointing means for assigning the point of the title.

According to a second aspect of the present invention, there is provided a method for extracting a title region from a document image, wherein a character string region is cut out in a rectangle from an input document image, and a point of title likeness is determined based on the attribute of the character string region. In a title area extracting method from a document image for performing a calculation and extracting a title, a character code in the character string area is recognized, and whether or not the certainty of the character code identification is equal to or greater than a certain threshold value is determined. A first step of determining and a second step of determining and extracting a title area based on the total value by adding points of the title likeness when the value is equal to or more than a predetermined threshold value in the first step. Including.

According to a third aspect of the present invention, there is provided a method for extracting a title region from a document image, wherein a character string region is cut out from the input document image in a rectangular shape, and a point of title-likeness is determined based on the attribute of the character string region. In a method for extracting a title from a document image for performing a calculation and extracting a title, a first step of performing character recognition in the character string area and calculating the number of characters in a character string rectangle at the time of character recognition;
Using the number of characters in the title of the document, comparing with the number of characters,
A second step of determining whether or not the number of character rectangles is within a predetermined value; and, in the second step, when the number of character rectangles is within a predetermined value, the points of the title-likeness are added, and the sum is calculated. A third step of determining and extracting a title area based on the value;
Is included.

According to a fourth aspect of the present invention, there is provided a method for extracting a title region from a document image, wherein a character string region is cut out from the input document image by a rectangle, and a point of title likeness is determined based on the attribute of the character string region. A first step of performing a natural language process on a recognition result of a character code in the character string area in a method of performing a calculation and extracting a title from a document image to extract a title; As a result, a second step of judging whether the area is a no-stop area and a point of title-likeness are added to the no-stop area in the second step,
A third step of determining and extracting a title area based on the total value.

According to a fifth aspect of the present invention, in the method for extracting a title region from a document image, a character string region is cut out from the input document image by a rectangle, and a point of title likeness is determined based on the attribute of the character string region. A first step of performing a natural language process on a recognition result of a character code in the character string area in a method of performing a calculation and extracting a title from a document image to extract a title; As a result, the statistical information dictionary of the endings appearing frequently in the title is compared with the character code string in the character string area, and it is determined whether or not the ending is a character string area including a ending that matches the frequent occurrence of the ending. And a third step of adding a point of title-likeness to the area of the second step and determining and extracting a title area based on the total value.

According to a sixth aspect of the present invention, in the method for extracting a title region from a document image, a character string region is cut out as a rectangle from the input document image, and a point of title likeness is determined based on the attribute of the character string region. In a title area extraction method from a document image for performing a calculation and extracting a title, a first step of performing font identification processing on the character string area, and a method of extracting characters based on a result of the font identification processing. A second step of determining a font style and determining whether the area is a character area using a specific font; and, for the character area using a specific font in the second step, a point of title-likeness is determined. And a third step of determining and extracting a title area based on the total value.

According to a seventh aspect of the present invention, there is provided a method for extracting a title region from a document image, wherein a character string region is cut out from the input document image in a rectangular shape, and the point of title likeness is determined based on the attribute of the character string region. In a method for extracting a title from a document image for performing a calculation and extracting a title, a first step of performing font identification processing on the character string area, and a font style based on a result of the font identification processing. A histogram of the font style of the entire document is created at the time of discrimination, and a second step of judging whether or not the character area uses a font style with a low appearance frequency, and a second step of judging the character area determined in the second step. On the other hand, a third step of adding the points of title-likeness and determining / extracting the title area based on the total value is included. That.

Further, in the method for extracting a title region from a document image according to claim 8, a character string region is cut out from the input document image by a rectangle, and a point of title likeness is determined based on the attribute of the character string region. A first step of calculating an aspect ratio of each character rectangle in the character string rectangle by performing a calculation and extracting a title from a document image; and determining whether the character is a double-width character based on the aspect ratio. And a third step of adding a title-like point to the character area determined to be the double-width character, and determining and extracting the title area based on the total value. .

According to a ninth aspect of the present invention, in the method for extracting a title area from a document image, a character string area is cut out from the input document image in a rectangular shape, and the point of the title likeness is determined based on the attribute of the character string area. In a method for extracting a title from a document image for performing a calculation and extracting a title, a first step of performing a character recognition process on the character string rectangle, and a step of recognizing a character other than a blank character by the character recognition process. A second step of calculating a total value of the horizontal width of the character rectangle (vertical width in vertical writing), a third step of determining whether the total value is substantially half of the character rectangular area, A fourth step of adding a title-like point to the character string area determined to be approximately half in the step, and determining and extracting the title area based on the total value.

In the method for extracting a title area from a document image according to a tenth aspect, the reference value used for judging whether or not the point likeness of the title can be added is learned in accordance with the input document format for each user. It is an optimal value that can be obtained and is variable and set.

Further, in the document search method according to the eleventh aspect, a first step of recognizing a document image, performing linguistic processing on the result, and extracting a keyword,
And the keywords extracted in the step (b).
A second step of writing the title extracted based on the title area extraction method from the document image described in any one of the above, and a document search is executed using the title written in the second step. And a third step.

[0018]

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a block diagram showing the configuration of a title area extracting apparatus and method for extracting a title area from a document image according to the present invention;

FIG. 1 is a block diagram showing a system configuration for performing a title area extracting process according to an embodiment of the present invention. In the figure, reference numeral 101 denotes an area identification unit as an area identification unit that extracts a character string area from a document image input from an image input device (not shown) such as a facsimile or an image scanner, and 102 denotes identification of the area identification unit 101. A character recognition unit 103 as a character recognition unit for performing character recognition based on the result, a font identification unit 103 as a font identification unit for performing font identification based on the identification result of the area identification unit 101, and a recognition unit 104 for the character recognition unit 102 Natural language analysis unit as natural language analysis means for analyzing the likelihood of a natural language title based on the character code obtained as a result, 1
Reference numeral 05 denotes a point-attaching unit as a point-attaching means for assigning a point of a title-likeness using the conventionally used centering, underlining, character rectangle size, and the like.

In the configuration shown in FIG. 1, when a document image is input from an image input device (not shown), preprocessing such as skew correction is performed, an area identification process is executed by an area identification unit 101, and a character string rectangle is formed. Information on the coordinate value and size of Next, using the result of the area identification processing by the area identification unit 101, character recognition by the character recognition unit 102 and font identification by the font identification unit 103 are performed.

In the character recognizing unit 102, a character code, a certainty factor, a coordinate value and a size of a character rectangle for each character are obtained as a point of the likeness of a title. In the font identification unit 103, the font type for each character is obtained as a point indicating the likeness of a title.

The character code obtained by the character recognizing unit 102 is also supplied to a natural language analyzing unit 104 and a natural language analyzing routine.
Gives the point of the title-likeness of the area where the speech is stopped. Furthermore, in natural language processing, the statistical information dictionary of the endings that appear frequently in the title is compared with the character code string in the character area, and the character string area that includes the ending that matches the frequent occurrence of the ending in the character string area has the title-likeness point. give.

Further, in addition to the above-mentioned points of the likelihood of points, the total points of the likelihood of the title are calculated using the conventionally used centering processing, underlining processing, the size of the character string rectangle, etc., and the title is identified. .

Next, a series of title extraction processing methods of the present invention will be described in order with reference to the flowcharts shown in FIGS. This title extraction process is described in FIG.
Depending on the configuration, a plurality of combinations, a single operation, or a selective operation can be performed.

FIG. 3 is a flowchart showing a first title extraction method according to the embodiment, in which a title-like point is added when the certainty of character code identification is equal to or greater than a certain threshold. Is shown. First, a document image is input from a document input device (not shown) (S301), and a character string area is identified by the area identification unit 101 (S302). Subsequently, the character code in the character string area is recognized, and it is determined whether or not the certainty factor of the character code identification is equal to or more than a predetermined threshold (S303). If it is determined that the value is equal to or more than the predetermined threshold value, the title-like points are added, and the title area is determined and extracted based on the total value (S304).

FIG. 4 is a flowchart showing a second title extracting method according to the embodiment. First, a document image is input from a document input device (not shown) (S401).
The character string region is identified by the region identification unit 101 (S40
2). Next, the number of characters in the character string rectangle is obtained at the time of character recognition (S403). The number of characters in the title of the document is compared with the number of characters (S404), and it is determined whether the number of character rectangles is within a predetermined value (S405). Here, if it is determined that the number of character rectangles is within the predetermined value, the title points are added, and the title area is determined and extracted based on the total value (S406).

That is, the number of characters in a character string rectangle is obtained at the time of character code recognition in the character string area, and the number of characters in the title of the document is compared as separate dictionary information using statistics. Gives the title point.

FIG. 5 is a flowchart showing a third title extracting method according to the embodiment. First, a document image is input from a document input device (not shown) (S501).
The character string area is identified by the area identification unit 101 (S50
2). Subsequently, natural language processing is performed on the recognition result of the character code in the character string area (S503), and it is determined whether or not the area is a predetermined area, for example, an area in which speech is stopped (S504). If it is determined that the area is the area of the predetermined item, the points of the title are added, and the title area is determined and extracted based on the total value (S505).

In the above-described language processing, a statistical information dictionary of endings frequently appearing in a title is compared with a character code string in a character area, and a character string area including a ending that coincides with a frequently appearing ending is added to the character string area. You may give the point of the title.

FIG. 6 is a flowchart showing a fourth title extracting method according to the embodiment. First, a document image is input from a document input device (not shown) (S601).
The character string region is identified by the region identification unit 101 (S60
2). Subsequently, font identification processing is performed (S603),
It is determined whether or not the area includes a predetermined font (font style) (S604). In other words, the font style of the character is determined, and the font style of the entire document is created when the font area is determined to be a character area or the font style is determined. It is determined whether it is a character area. Then, when it is determined that these areas are present, points of the title likeness are added, and the title area is determined and extracted based on the total value (S605).

FIG. 7 is a flowchart showing a fifth title extracting method according to the embodiment. First, a document image is input from a document input device (not shown) (S701),
The character string area is identified by the area identification unit 101 (S70
2). Subsequently, the aspect ratio of each character rectangle in the character string rectangle is obtained (S703), and the aspect ratio is horizontal: vertical = 2: 1.
It is determined whether or not the character rectangle having a value close to is occupying a certain percentage or more of the number of character rectangles in the character string rectangle (S7).
04). Here, if the ratio occupies a certain ratio or more, the title-like points are added, and the title area is determined and extracted based on the total value (S705).

FIG. 8 is a flowchart showing a sixth title extracting method according to the embodiment. First, a document image is input from a document input device (not shown) (S801).
The character string area is identified by the area identification unit 101 (S80
2). Subsequently, character recognition processing is performed (S803), and the total value of the horizontal width (vertical width in vertical writing) of each character rectangle recognized as a character other than a blank character by the character recognition processing is obtained (S80).
4). Then, it is determined whether or not the total value is substantially half of the character rectangular area (S805). Here, if the total value is substantially half of the character rectangular area, the points of the title likeness are added, and the title area is determined and extracted based on the total value (S806).

By the way, without fixing the threshold value required in the above-described embodiment, learning is performed in correspondence with the document input by each user, and the optimum threshold value is set for the document format used by each user. May be variably obtained and changed / set from the initial value.

Further, based on the temporary points obtained as described above, double-width characters and uniform layout are determined by a quadratic combination as shown in FIG. It is also possible to give.

In addition, the double-width character is determined by calculating the aspect ratio of the character rectangular area using the size of each character rectangular area in the character string rectangle obtained at the time of character code recognition. Is given a title-like point for the character string area using.

The character density in the rectangle is calculated using the size of the character rectangle region and the character string region to which it belongs, and the number of characters in the character string rectangle obtained at the time of character code recognition.
Equal allocation determination is performed based on the value. Then, a point of the title-likeness is given to the character string area determined to be equally allocated.

Incidentally, it is also feasible to perform information retrieval using the above-described title area extraction method. FIG.
9 is a flowchart illustrating an information search method according to the embodiment. First, a document image is recognized (S901), and language processing is performed on the result to extract a keyword (SS902). Further, the extracted keyword and the title extracted by the above-described title area extraction method are written together (S903), and a document search is executed using the combined title (S904). Thereby, the convenience at the time of search is improved.

[0038]

As described above, according to the apparatus for extracting a title area from a document image according to the present invention (claim 1), there is provided an area identifying means for extracting a character string area from an input document image by a rectangle. Then, based on the attribute of the character string area, a point calculation of the likelihood of a title is executed, and when extracting the title, character recognition within the character string rectangle is performed on the character string rectangle cut out by the area identification means. A character recognition unit, a font identification unit that identifies a font of each character in the character string rectangle cut out by the region identification unit, and a character code obtained from a recognition result of the character recognition unit. Language analysis means for analyzing the likelihood of a natural language title, and the character string rectangle extracted by the area identification means, using centering, underlining, the size of the character rectangle, etc. A point attaching means for assigning points is provided, and a character string area having a large number of points is automatically extracted as a title because a characteristic unique to a title is used as the point attaching without depending on a specific document format. In addition, it is possible to provide a device that improves the accuracy of title extraction and the convenience at the time of document search.

According to the method for extracting a title area from a document image according to the present invention (claim 2), a character code in a character string area is recognized, and the certainty of character code identification is equal to or greater than a certain threshold value. Judgment is made, and if it is more than a certain threshold, points like a title are added, and the title area is determined and extracted based on the total value. This makes it possible to improve the accuracy of title extraction and the convenience in document search.

According to the method for extracting a title area from a document image according to the present invention (claim 3), character recognition in a character string area is executed, and the number of characters in a character string rectangle is obtained at the time of character recognition. Using the number of characters in the title of the document, compare with the above number of characters to determine whether the number of character rectangles is within a predetermined value. If the number of character rectangles is within a predetermined value, add the point of title-likeness to the corresponding character string area. Since the title area is determined and extracted based on the sum, the character string area having a large number of points can be automatically extracted as a title, and the accuracy of title extraction and the convenience of document search are improved. be able to.

According to the method for extracting a title area from a document image according to the present invention (claim 4), natural language processing is performed on the recognition result of the character code in the character string area,
As a result, it is determined whether or not the area is in a no-stop area, points of the title-likeness are added to the no-stop area, and the title area is determined and extracted based on the total value. It is possible to automatically extract a large number of character string regions as titles, and improve the accuracy of title extraction and the convenience in document search.

According to the method for extracting a title area from a document image according to the present invention (claim 5), natural language processing is performed on a result of recognition of a character code in a character string area,
As a result, the statistical information dictionary of the endings appearing frequently in the title is compared with the character code string in the character string area, and it is determined whether or not the ending is a character string area that matches the ending with the high frequent occurrence. The title-like points are added to the area, and the title area is determined and extracted based on the total value. Therefore, it is possible to automatically extract a character string area having a large number of points as a title. Convenience at the time of document search can be improved.

According to the method for extracting a title region from a document image according to the present invention (claim 6), font identification processing is performed on a character string region, and based on the result,
Determines the font style of the character, determines whether it is a character area using a specific font, adds a point of title-likeness to the character area using a specific font, and calculates the title area based on the total value. In order to determine and extract a character string, it is possible to automatically extract a character string region having a large number of points as a title, and to improve the accuracy of title extraction and the convenience in document search.

According to the method for extracting a title area from a document image according to the present invention (claim 7), font identification processing is performed on a character string area, and based on the result,
A histogram of the font style of the entire document is created at the time of determining the font style, and it is determined whether or not the character area uses a font style with a low frequency of appearance.
The title area is added to the determined character area, and the title area is determined and extracted based on the total value. Therefore, it is possible to automatically extract a character string area having a large number of points as a title, and to extract the title. Accuracy and convenience at the time of document search can be improved.

According to the method for extracting a title region from a document image according to the present invention (claim 8), the aspect ratio of each character rectangle in the character string rectangle is obtained, and based on the aspect ratio, double-width characters are obtained. The character area that has a large number of points is automatically extracted as the title because the title area is added to the character area determined to be double-width characters, and the title area is determined and extracted based on the total value. This makes it possible to improve the accuracy of title extraction and the convenience in document search.

According to the method for extracting a title region from a document image according to the present invention (claim 9), a character recognition process is performed on a character string rectangle, and each character other than a blank character is recognized by the character recognition process. Calculate the total value of the width of the character rectangle (vertical width when writing vertically), determine whether the total value is almost half of the character rectangle area, and assign the title to the character string area determined to be almost half. Add the points of likeness,
To determine and extract the title area based on the total value,
It is possible to automatically extract a character string region having a large number of points as a title, and improve the accuracy of title extraction and the convenience in document search.

According to the method of extracting a title area from a book image according to the present invention (claim 10), the title area is extracted from a document image according to any one of claims 2 to 9. By changing and setting the reference value used to determine whether or not the likelihood point addition is possible using the optimum value obtained by learning according to the input document format for each user, more accurate automatic title extraction is realized.

Further, according to the document search method of the present invention (claim 11), a keyword extracted by performing character recognition on a document image and performing linguistic processing on the result is stored in the document search method. In order to execute a document search using a title extracted from a document image described in any one of them based on a title area extraction method based on the title area extraction method, that is, a more accurate title, a document search is performed. The convenience at the time is improved.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a system configuration for performing a title region extraction process according to an embodiment of the present invention.

FIG. 2 is a block diagram showing secondary points obtained among points of the likelihood of a title used in the title area extraction processing according to the embodiment of the present invention;

FIG. 3 is a flowchart illustrating a first title extraction method according to the embodiment of the present invention.

FIG. 4 is a flowchart illustrating a second title extraction method according to the embodiment of the present invention.

FIG. 5 is a flowchart showing a third title extracting method according to the embodiment of the present invention.

FIG. 6 is a flowchart showing a fourth title extracting method according to the embodiment of the present invention.

FIG. 7 is a flowchart showing a fifth title extracting method according to the embodiment of the present invention.

FIG. 8 is a flowchart showing a sixth title extracting method according to the embodiment of the present invention.

FIG. 9 is a flowchart showing an information search method according to the embodiment of the present invention.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 101 Area identification part 102 Character recognition part 103 Font identification part 104 Natural language analysis part 105 Pointing part

Claims (11)

[Claims] 1. An image processing apparatus comprising: an area identifying means for extracting a character string area from a document image input from an image input device in a rectangular form; performing point calculation of title likelihood based on an attribute of the character string area; A title region extraction unit for performing character recognition on the character string rectangle extracted by the region identification unit, and a character string extracted by the region identification unit. Font identification means for identifying a font for each character in the character string rectangle with respect to the rectangle, and natural language analysis means for analyzing the likelihood of a natural language title based on the character code obtained by the recognition result of the character recognition means And, for the character string rectangle cut out by the area identification means, pointing of the likeness of the title using centering, underlining, the size of the character rectangle, and the like. A title region extracting device for extracting a title region from a document image, the device comprising: 2. A method for extracting a title region from a document image, wherein a character string region is cut out in a rectangle from an input document image, a point calculation of a title likeness is performed based on an attribute of the character string region, and a title is extracted. A first step of recognizing a character code in the character string area and determining whether or not the certainty of character code identification is equal to or greater than a predetermined threshold value;
And a second step of adding a point of title-likeness when the value is equal to or greater than a predetermined threshold value in the first step, and determining and extracting a title area based on the total value. Title area extraction method from the document image to be described. 3. A method for extracting a title region from a document image in which a character string region is cut out in a rectangle from an input document image, a point calculation of a title likeness is performed based on the attribute of the character string region, and a title is extracted. A first step of performing character recognition in the character string area and determining the number of characters in the character string rectangle at the time of character recognition; and comparing the character number with the character number by using the number of characters in the title of the document. A second step of determining whether the value is within a value,
A step of adding title-like points when the number of character rectangles is within a predetermined value, and determining / extracting a title area based on the total value. Method of extracting the title area from. 4. A method for extracting a title area from a document image in which a character string area is cut out in a rectangle from an input document image, a point calculation of a title likeness is performed based on the attribute of the character string area, and a title is extracted. A first step of performing a natural language process on a recognition result of a character code in the character string area, and a second step of determining whether or not the area is a speech stop area as a result of the first step. And a third step of adding a point of title-likeness to the area stopped in the second step, and determining and extracting the title area based on the total value. Title area extraction method from the document image to be described. 5. A method for extracting a title area from a document image in which a character string area is cut out in a rectangle from an input document image, a point calculation of a title likeness is performed based on the attribute of the character string area, and a title is extracted. A first step of performing a natural language process on the recognition result of the character code in the character string area, and a statistical information dictionary of endings frequently appearing in the title as a result of the first step; To determine whether or not it is a character string area that includes a suffix that matches the high-frequency ending.
And a third step of adding title-like points to the area of the second step, and determining and extracting a title area based on the total value. Title area extraction method. 6. A method for extracting a title area from a document image in which a character string area is cut out in a rectangle from an input document image, a point calculation of a title likeness is performed based on the attribute of the character string area, and a title is extracted. A first step of performing font identification processing on the character string area, and determining a font style of the character based on a result of the font identification processing to determine whether the character area uses a specific font. A second step of determining
For the character area that uses a specific font in the process of
A third step of adding a point of title-likeness and determining and extracting a title area based on the total value thereof. 7. A method for extracting a title area from a document image, wherein a character string area is cut out in a rectangle from an input document image, a point calculation of a title likeness is performed based on the attribute of the character string area, and a title is extracted. A first step of executing a font identification process on the character string region; and, based on the result of the font identification process, creating a font style histogram of the entire document at the time of font style determination. A second step of determining whether or not the character area uses a small font style; and adding a point of title-likeness to the character area determined in the second step, and determining a title area based on the total value. A method of extracting a title area from a document image, the method including: a third step of extracting. 8. A method for extracting a title area from a document image in which a character string area is cut out in a rectangle from an input document image, a point calculation of a title likeness is performed based on the attribute of the character string area, and a title is extracted. A first step of determining an aspect ratio of each character rectangle in the character string rectangle, a second step of determining whether the character is a double-width character based on the aspect ratio, and a character determined to be the double-width character Adding a point of title-likeness to the area and determining / extracting the title area based on the total value. 3. A method of extracting a title area from a document image. 9. A method for extracting a title area from a document image in which a character string area is cut out in a rectangle from an input document image, a point calculation of a title likeness is performed based on the attribute of the character string area, and a title is extracted. A first step of performing a character recognition process on the character string rectangle, and obtaining a total value of the horizontal width (vertical width in vertical writing) of each character rectangle recognized by the character recognition process other than a blank character. Step 2, a third step of determining whether the total value is substantially half of the character rectangular area, and a character string area determined to be substantially half in the third step. And a fourth step of determining and extracting a title area based on the total value of the points. 10. A reference value used for determining whether or not point addition of title likeness is possible is an optimal value obtained by learning according to an input document format for each user, and is variably set. A method for extracting a title area from a document image according to any one of 2 to 9. 11. A first step of recognizing a document image, performing language processing on the result, and extracting a keyword,
A second step of writing together the keyword extracted in the first step and a title extracted based on the title area extraction method from a document image according to any one of claims 2 to 10; A third step of performing a document search using the title described in the second step.