JPH09305701A - Form recognition method - Google Patents

Abstract

(57) [Abstract] [Problem] To propose a highly accurate form recognition method for a reading object having various kinds of forms. It is also to propose a type identification method. It is also to provide a method for extracting underlines described in a form. SOLUTION: Ruled line frames 204, 206 from a form image 200
The character line 212 is extracted, and the character identification result is collated with the word dictionary to correct the character identification error. The type of form is identified based on the characteristics of the table and the form name and item name obtained by matching. Character lines and ruled lines are extracted from the form image, the ruled lines that make up the frame are removed from the extracted ruled lines, and the underlines are extracted by comparing the arrangement of the remaining ruled lines and the character lines. [Effect] It is possible to identify the form type with high accuracy even for non-standard forms such as registered notices, and to extract underlines with high accuracy without making mistakes with strokes in characters. .

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a form, in particular, various forms in which registration information regarding real estate is described, and more particularly, to a form recognition method for reading character data from a registered notice and automatically inputting it.

[0002]

2. Description of the Related Art The following is an example of a conventional technique for identifying the type of a form. The first is a method of identifying the type of the form by reading the ID number representing the type of the form described in the same position for all types of forms. The second is a method of identifying the type of form by identifying the structure of the frame when the structure of the frame differs for each type of form. This example is described in JP-A-7-1414.
No. 62 publication.

[0003]

[Problems to be Solved by the Invention] There are currently seven types of registered notices regarding real estate. These forms are used to enter data for taxation of real estate,
Since this notification does not include an ID number that identifies the type of the form, the conventional method of identifying the form by reading the ID number cannot be used. Furthermore, since these forms are atypical forms with different frame shapes even if they are of the same type, it is not possible to use the conventional method of identifying a form from the frame structure. Further, when the conventional method for identifying a form by reading the characters in the title part is used, there is a problem that the form identification accuracy greatly depends on the character recognition accuracy. The form name of the registered notice is “land registered notice regarding rights”, “building registered notice regarding rights (general)”, “building registered notice regarding rights (proprietary)”, “land registration regarding labeling” There are seven types: “Notification of completed building”, “Notification of registered building related to display (general)”, “Notification of registered building related to display (one building)”, “Notification of registered building related to display (proprietary)”. Of these, the "registered building registration notice for display (general)" and the "registered building registration notice for display (one building)" differ by only one character, so
The identification accuracy for these two types may be low.

Therefore, a first object of the present invention is to propose a form recognition means having a highly accurate form identification means for reading objects having various kinds of forms.

In the conventional underline detection method, since ruled lines other than the frame line are used as underlines, there is a possibility that noise components such as horizontal strokes of characters may be erroneously extracted as underlines. Therefore, a second object of the present invention is to propose a form recognition means having a highly accurate underline detection means.

[0006]

According to a first aspect, the present invention is a method of recognizing registration information for inputting a front surface image of a registered notice and reading characters, wherein a character line is read from the image of the registered notice. Registered from the character line extracting means for extracting the character line, the character line selecting means for selecting the character line of the form name from the positional relationship between the extracted plurality of character lines and the frame, and the character identifying means for reading the character line of the form name. First form identification means for identifying the type of notice, ruled line extraction means for extracting ruled lines from the image of the registered notice, table feature extraction means for extracting table features from the extracted ruled lines, and table features Second form identifying means for identifying the type of registered notice from, a character line extracting means for extracting a character line from the image of the registered notice, a character identifying means for reading the extracted character line, and a reading result Item name to select the item name of the form from A form for identifying the type of registered notice by including selection means and third form identifying means for identifying the type of registered notice based on the combination of item names, and combining the results of the three means for identifying the type of registered notice. Provide a recognition method.

According to a second aspect, the present invention is a method of recognizing registration information for inputting a surface image of a registered notice and reading characters, wherein characters for extracting character lines and ruled lines from the image of the registered notice. A line extracting unit, a ruled line type determining unit that distinguishes a frame ruled line from a ruled line that is not a frame ruled line based on the extracted ruled lines, a unit that detects a character line in a frame that includes a ruled line that is not a frame ruled line, and a character line in the frame Provided is a form recognition method including an underline detection unit that determines whether a ruled line that is not a frame ruled line in the frame is an underline based on a positional relationship with a ruled line that is not a frame ruled line in the frame.

[0008]

DESCRIPTION OF THE PREFERRED EMBODIMENTS One embodiment of the present invention will be described below in detail. The present invention is not limited to this.

FIG. 1 is a block diagram of a registration information system which is an embodiment of the present invention. The recognition unit 101 that recognizes the registration information and the correction unit 105 that corrects the recognition result are connected by the network 104, and the input center 111 can perform the recognition and the correction in parallel. The process of processing is as follows.
Enter the image of. Next, the recognition computer 103 recognizes characters and ruled lines, and the correction computer 106 confirms correction of the recognition result. It also checks the collation with a dictionary or code table and outputs the code data. The recognition result is stored in the registration information database 109 connected to the host computer 108 in the remote computing center 110 via the communication control computer 107. Correction unit 10
In 5, the part of the recognition result is used to access the registration information database 109 and read the registered registration information.
The read registration information and a part of the recognition result are collated to test whether there is any contradiction.

FIG. 2 is a block diagram showing the process of registration information recognition. The recognition unit 101 reads the form image and outputs a reduced image 248, frame coordinates 250, underline coordinates 252, character line coordinates 254, form type 256, recognition result lattice 258, and character coordinates 260 to the correction unit 105. In the correction unit 105, the operator corrects the recognition result based on these input data. In the image input unit 200, the image on the front surface of the form is binarized and input.

The input image is output to the image reduction unit 202 and the character line image extraction unit 218. The image reduction unit 202 reduces the form image and outputs a reduced image 248 in order to speed up the subsequent processing. In the reduction processing, OR processing is performed for each pixel so that a thin ruled line does not become faint after the reduction. The ruled line extraction unit 204 extracts ruled lines of solid lines and dotted lines from the reduced image. The solid line is extracted based on the continuous connection of black pixels. The dotted lines are extracted based on the arrangement of the circumscribed rectangles of the connected components of black pixels and the size constraint conditions. The frame extraction unit 206 obtains a frame surrounded by the ruled lines on all sides from the ruled line extracted in 204, and outputs the vertex coordinates 250 of the frame. The table feature extraction unit 208 extracts the feature amount of the table, which is a set of frames, from the information of the frames extracted in 206. The feature amount is the number of vertical and horizontal ruled lines, the connection relationship between ruled lines, the positional relationship between frames, and the like.

On the other hand, the character line extraction unit 206 extracts a character line, which is a set of characters, from the reduced image output from 202. Here, a character line is generated by fusing the circumscribed rectangles that are estimated to be a sequence of characters based on the vertex coordinates of the circumscribed rectangles of the connected components that are estimated to be characters among the connected components of black pixels. To do. In the line-frame correspondence unit 214, 2
By comparing the vertex coordinates of the character line extracted in 12 and the vertex coordinates of the frame extracted in 206, it is determined in which frame each character line exists or is outside the frame, and it is included in each frame. The vertex coordinates of the character line and the vertex coordinates 254 of the character line outside the frame are output. Further, in the underline extraction unit 216, 2
Underlines are extracted based on the ruled line coordinates extracted in 04, the vertex coordinates of the frame extracted in 206, and the character line coordinates in the frame extracted in 214, and the underline coordinates 252 are output. Further, the character line image extraction unit 218 cuts out the image of the character line portion from the image input in 200 based on the character line coordinates extracted in 214. The character cutout / character identification unit 220 includes a character cutout unit 222 and a character identification unit 224.
Cooperate with each other to cut out characters one by one and output the character coordinates 260. Further, in the character identification unit 224, the identification dictionary 2 is added to the extracted image pattern of one character.
26 is used to identify the character. The form name collation unit 228 inputs the character identification result output from the character identification unit 224, and the word collation unit 230 collates the result with the form name word stored in the form name dictionary 232 to obtain the recognition result of the form name. Correct the mistake and obtain the form name.

The words stored in the form name dictionary 232 are the form names to be recognized. The form name to be recognized is known in advance, and the form name corresponds to the type of the form one-to-one. Further, in the item matching unit 234, the character recognition result which has not been matched in 228 is input, and the word matching unit 236 matches the character name with the item name word stored in the item dictionary 238 to correct the error in the recognition result of the item name. Correct and find the item name. The words in the item dictionary 238 are items described in the form to be recognized. Content collation unit 240
Then, the character recognition result that has not been matched in 234 is input, and the word matching unit 242 corrects the error in the recognition result regarding the content by matching with the content word stored in the content dictionary 244. Here, the “content” refers to the content described for the item name in the form. For example, the contents for the item "ground" include "home" and "park". The words stored in the content dictionary 244 are words that are predetermined to be used among the words that describe the content described in the form to be recognized.
Recognition result lattice 258 output as a result of processing 240
Is a sequence of candidate characters, which are the result of the character identification processing, arranged in descending order of similarity for each character. This character identification result is corrected for errors by form name matching, item matching, and content matching.

On the other hand, the form identification unit 246 inputs the output results of the table feature extraction unit 208, the form name comparison unit 228, and the item name comparison unit 234, and identifies the form type from the table feature, the form name, and the item name. , Form type 256 is output.

FIG. 3 is a diagram showing a processing flow of registration information recognition shown in FIG. An image is input in step 300, and the image is reduced in step 302. Next, in step 304, ruled lines are extracted from the image, and in step 306
Extract the frame from the ruled line with. Further, in step 308, the characteristics of the table are extracted. In step 310, character lines are extracted from the reduced image, and in step 312, the extracted lines are associated with the frames. In step 314, the underline is extracted from the coordinates of the ruled line, the frame, and the character line. Further, in step 316, only the image of the character line portion is extracted from the form image based on the coordinate value of the character line. In step 318, the character line image is divided into images for each character, and step 3
Character identification is performed on the image pattern cut out in 20. In step 322, the character identification result is collated with the word of the form name to identify the form name. In step 324, the character identification result is collated with the word of the item name to identify the item name. In step 326, the content is identified by matching the character identification result with the word of the content. In step 328, the characteristics of the table as the processing result of step 308 and step 32
The type of the form is identified from the form name which is the processing result of No. 2 and the item name which is the processing result of step 324. Step 33
At 0, the result obtained by the processing from 300 to 328 is output.

FIG. 4 is a diagram simply showing an image of a registered notice as a recognition target for the purpose of explanation. In the example of the form image 400, the form name “Building registration notice regarding rights (proprietary)” 401 is described. Horizontal ruled lines 402,
404, 406, 408 and vertical ruled lines 410, 412, 41
4, 416 are printed. Further, as items, there are “code” 418, “location” 420, and “ground level” 422.
The contents of "sign" are "1" (424) and "2" (4
26), the content of “location” is described in 428 and 430 as “280-chome, Higashikoigakubo, Kokubunji-shi”.
As the contents of "ground", "residential area" (432) and "park" (434) are described. In addition, content 424
Underlines 436 and 43 under "1", 428, "280-chome, Higashi Koigakubo, Kokubunji," and 432 "residential land," respectively.
8,440 are printed.

FIG. 5 shows the ruled line extraction processing result of step 304 of FIG. 3 for the form image of FIG.
500 in (a) is the extraction result of the horizontal ruled line, and 5 in (b)
Reference numeral 20 is the result of vertical ruled line extraction. In (a), 502 to 508 are extracted as ruled lines corresponding to the horizontal ruled lines 402 to 408 of FIG. 4, respectively. Underline 436
As underlines corresponding to 438 and 440, respectively, 51
0,514,516 are extracted. 512 and 518
Is extracted as a ruled line by connecting horizontal strokes of "Ichito Koi". The phenomenon in which the horizontal strokes that are separated from each other are connected is that black pixels that are close to each other are connected by contracting and expanding the black pixels in the horizontal direction when extracting the horizontal ruled line. Further, in (b), 522 to 528 are extracted as ruled lines corresponding to the vertical ruled lines 410 to 416 of FIG. 4, respectively.

FIG. 6 shows the result of frame extraction processing in step 306 of FIG. 3 for the form image of FIG. 6
00 is a frame extraction result. Nine frames 602 to 618 are extracted.

FIG. 7 shows the character line extraction processing result of step 310 of FIG. 3 for the form image of FIG. Reference numeral 700 is a character line extraction result. Character line 40 in FIG.
The circumscribing rectangles of the character lines 702 to 720 are extracted for the character lines 1,418, 420, 422, 424 to 434, respectively.

FIG. 8 is a processing flow relating to the character line extraction processing in step 314 of FIG. Ruled line extraction processing 30
4. In step 800, ruled lines that do not form a frame are extracted using the results of the frame extraction process 306 and the character line extraction process 310. In step 802, the following processing is repeated for the number of ruled lines extracted in step 800. In step 804, the coordinates of the character line and the coordinates of the ruled line are compared. The comparison method will be described with reference to FIGS. 9 and 10. In step 806, it is determined whether the compared value satisfies the standard. If the reference value is satisfied, then in step 808, the ruled line to be compared is underlined. Note that the above step 808
Regarding the two underlines extracted in (1), if the end points are separated from each other by a minute interval and are present on the extension line, it can be regarded as one underline. Further, if the length of the underline extracted in step 808 is equal to or less than the reference value, it may not be considered as an underline.

FIG. 9 is an example of a form frame for explaining the processing flow of FIG. Horizontal ruled lines 900 and 902, vertical ruled lines 904 and 906, character lines 908, and underlines 910 are printed.

FIG. 10 shows the result of extracting ruled lines and character lines from the example of FIG. The underline determination will be described with reference to this figure. In the underline determination processing, a ruled line that is located below a character line and has a length substantially the same as that of the character line is determined to be an underline. In FIG. 10, 1007 is an area in which characters are printed, and 1008 is a circumscribed rectangle of 1007. The ruled lines 900 to 910 in FIG. 9 are extracted as 1000 to 1010, respectively. Further, 1012 is a horizontal stroke of a character extracted as a ruled line. From the extracted ruled lines, 1010 and 1012 are extracted as ruled lines that do not form a frame.
Hereinafter, a case where it is determined that the line is underlined will be described by taking 1010 as an example, and a case where the line is not determined as underline is described by taking 1012 as an example.

The judgment is made for 1010 in FIG. First, the difference d11 (1014) between the y coordinate of the lower end of the ruled line and the y coordinate of the lower end of the character line is obtained. Next, the difference d12 (1016) between the y coordinate of the upper edge of the ruled line and the y coordinate of the upper edge of the character line is obtained. Furthermore, the length L1 of the ruled line in the x direction (1018)
And the length Lc (1020) of the character line in the x direction are calculated. This value is compared with the reference values α, β, γ1, γ2.
If d11 is below the character line and less than α, d12 is β or more, and L1-Lc is γ1 or more and γ2 or less, this ruled line is underlined. Α, which is the criterion for the above processing,
The values of β, γ1 and γ2 can be empirically obtained.

For example, the value of α can be used if the distance between the character line and the underline is constant. If it is not constant, 1/2 of the difference between the height of the frame and the height of the character can be used. For β, the sum of these two values can be used if the distance between the lower end of the character line and the underline and the height of the character are constant. As for the values of γ1 and γ2, a value obtained by multiplying the character width by (-1) can be used for γ1, and a character width or the like can be used for γ2 in consideration of a margin of about one character. The above α, β, γ1,
In setting the value of γ2, the value can be set with a margin in order to provide robustness against the inclination of the form, the blurring of lines, and the like. In addition, if a negative value is allowed for the allowable value of d11, it is possible to deal with the case where the underline overlaps a character.

Next, determination is made for 1012 in FIG. First, a difference d21 (1022) between the y-coordinate of the lower end of the ruled line and the y-coordinate of the lower end of the character line is obtained. Next, the difference d22 (102) between the y coordinate of the upper edge of the ruled line and the y coordinate of the upper edge of the character line.
Find 4). Furthermore, the length L2 (10
26) and the length Lc (1020) of the character line in the x direction are calculated. When these values are compared with the above α, β, γ1, γ2, d21 has a large negative value and d22 has a value smaller than β, so it is determined that the line is not underlined.

The d11 and d12 used here may be normalized by the height of the character or the height of the frame. Further, the ratio may be compared instead of the difference between L1 and Lc. α, β, γ
The values of 1 and γ2 are set according to the definition of the comparison target.

Further, here, the difference 1014 between the y-coordinate of the lower end of the ruled line and the y-coordinate of the lower end of the character line, and the y-value of the upper end of the ruled line.
The difference between the coordinates and the y-coordinate of the top of the character line 1016, the ruled line x
Direction length (1018) and character line length in the x direction (10
Although three evaluation values of the difference from 20) were used, only one or two of them may be used as necessary.

FIG. 11 shows an example in which the character coordinates are used instead of the character line coordinates in the step 314 underline extraction processing of FIG. By comparing the ruled line 1108 that does not form a frame with the circumscribing rectangle 1112 of the character using the determination criteria described in FIG. 10, it can be determined that 1108 is an underline. Further, by comparing the ruled line 1110 that does not form a frame with the circumscribing rectangle 1114 of the character, it can be determined that 1110 is not an underline.

FIG. 12 shows an example in which underlines are printed only on some characters in a character line. In the frame 1200,
A character line 1202 and an underline 1204 are described. FIG.
If the method 1 is used, it is possible to determine that the underline is printed only on the character “1, 280” in the character line.

FIG. 13 is another processing flow relating to the character line extraction processing in step 314 of FIG. In the registered notice, a plurality of underlines are often present on the same line like 436, 438, and 440 in FIG. On the other hand, the underline 436
Is short, the length is the same as the horizontal stroke in the character, so there is a possibility that it will be missed when the ruled line is extracted. The purpose of this process is to correctly extract short underlines that may be missed during ruled line extraction. Therefore, first, a long underline is extracted, and a ruled line that is an extension of this underline is determined as an underline.

Each step in FIG. 13 will be described below. In step 1300, only long underlines are extracted.
This processing can be realized using the processing shown in FIG.
In step 1302, run length data that does not form a frame line is extracted from the horizontal run length data.
In step 1304, the processes of steps 1306 and 1308 are repeated for the number of extracted run length data. In step 1306, it is determined whether the target run length data is on the extension of the underline. If it is on the extension line, it is extracted in step 1308 as run length data forming the underline. Step 13
In step 10, the ruled line composed of the run length data determined to form the underline in step 1308 is extracted as the underline. It should be noted that the two underlines extracted in step 1310 may be regarded as one underline if the endpoints are separated from each other by a minute distance and are present on the extension line. In addition, the above step 131
If the length of the underline extracted at 0 is less than the reference value,
You can choose not to consider it as an underline.

FIG. 14 is an example of a form frame for explaining the processing flow of FIG. Horizontal ruled lines 1400 and 140
2, vertical ruled lines 1404 to 1410, underlines 1412 to 1
416, character lines 1418 to 1422 are printed.

FIG. 15 is a result of extracting run length data in the horizontal direction and long underlines that do not form a frame from the image of FIG. 1500 is a long underline extracted in step 1300 of FIG. Among the connected components of the run length data in the horizontal direction, 1502 and 1504 are within the permissible range w (1510) from the extension line 1508 of 1500, and thus are determined to be underlines. Since 1506 is outside w, it is determined that there is no underline.

FIG. 16 is another processing flow relating to the character line extraction processing in step 314 of FIG. In this process, the underline is extracted using a histogram created by projecting the value of the length of the horizontal run length data that does not form a frame from the midpoint of the run in the tilt direction. Hereinafter, each step of FIG. 16 will be described. In step 1600, run length data that does not form a frame line is extracted from the run length data in the horizontal direction. In step 1602, the length value of the extracted run length data is projected from the midpoint of the run in the tilt direction to create a histogram. In step 1604, the number of peaks in the histogram is set to step 1
The processing of 606 and step 1608 is repeated. In step 1606, it is determined whether the projection value is equal to or larger than the reference value. If it is greater than or equal to the reference value, it is determined that the run length data projected in step 1608 forms an underline.
In step 1610, the underline is extracted from the run length data determined to form the underline in step 1608. Note that the two extracted in step 1610 above
Regarding the underline, if the end points are separated from each other by a minute distance and are present on the extension line, it can be regarded as one underline. Further, if the length of the underline extracted in step 1610 is equal to or smaller than the reference value, it may not be considered as an underline.

FIG. 17 shows the result of extracting the run length data in the horizontal direction which does not form a frame from the image of FIG. 14 and creating a histogram. 1700 to 1706 are shown in FIG.
Is the connected component of the run length data in the horizontal direction extracted in step 1600. Histograms 1708 and 17
10 is the result projected in step 1602. In step 1606, for 1708 and 1710, the area within the allowable range w (1712) is compared with the reference value. In this case, 1708 is the reference value or more, 1710
Is less than the reference value, 1700, 1702, 1
It can be determined that 704 is an underline and 1706 is not an underline.

FIG. 18 is a process flow relating to the form identification process of step 328 of FIG. In step 308, the feature amount of the table is extracted. In step 322, the word matching result of the form name is obtained. In step 324, the word matching result of the item name is obtained. In step 1800, 308,3
Using the form types respectively derived from the results of 22 and 324, the form type is identified by majority vote.

The characteristics of the table extracted in step 308 include connection relation of ruled lines, number of frames, arrangement relation of frames, number of vertical ruled lines, number of horizontal ruled lines, and the like. If the connection relationship of ruled lines differs depending on the type of form, it is disclosed in Japanese Patent Laid-Open No. 7-141462.
The type of form can be specified by using the technology described in the publication.

[0038]

[Table 1]

In Table 1, as an example of the characteristics of the table extracted in step 308, the number of vertical solid line ruled lines of the registered notice to be recognized is shown. From this, it can be seen that the vertical solid line ruled line is any of 7, 8, 10, 11, 12, and 16. Of these, the types of forms are uniquely determined except for the cases of 8 and 10. Even in the case of 8 and 10, candidates for the form type can be mentioned.

As for the word of the form name to be checked in step 322, all the form names may be registered as one word, and “right”, “display”, “building”, “land”, “general”,
Only characteristic words such as "proprietary" and "one building" may be registered.

[0041]

[Table 2]

Table 2 shows a part of the item names checked in step 308. From Table 2,
There are item names that are common to multiple forms, such as "location" and "place," and item names that are unique to the form, such as "land area", "building number of one building", "building", and "table". . Even for a type of form that does not have an item name unique to the form, a combination of multiple items is used to determine the existence, and thus a “building registration notification for display (general)” and a “building registration notification for display (proprietary) ) ”Can be identified. For example, if the item "floor area" exists and the item "building number of one building" does not exist, it can be identified as "notification of building registration regarding rights (general)".

In step 1800, in step 308,
The types of forms are identified by integrating the results of 322 and 324. As a means of integration, a majority vote of the above three results can be used.

In step 1800, 308, 32
Even if the form type cannot be uniquely identified in each of steps 2 and 324, the form type can be identified by mutually complementing the processing results of each step.
For example, when eight vertical solid ruled lines are extracted in step 308, from Table 1, the types of forms are “land registration notice regarding display”, “building registration notice regarding display (general)”, There are three types of “building registration notification regarding display (proprietary)”. But step 3
If the item name “table” is extracted in 24, it can be uniquely determined to be the “land registration notice regarding display”.

In step 1800, 30
It is also possible to use only two instead of using the results of the three steps of 8,322,324.

In step 1800, 30
Rather than treating the results of steps 8, 322, and 324 equally, identify the form from the result obtained in one step, and use the results obtained in the other steps to verify the result of form identification. You can also

FIG. 19 is a block diagram of a registration information system which is an embodiment of the present invention. The configuration of 101 to 109 is the same as in FIG. The sorter 1900 sorts the forms 100 in the order of the priority of the registered notices based on the result recognized by the recognition unit 101 and corrected by the correction unit 105. Two examples of sorting will be given below. The first is to sort the streets from the letters corresponding to the location and lot number in the order of chome, street number, and issue. Second, sort by creation date and number.
Further, the object to be sorted may be the form of the registered notice or the data of the recognition result.

[0048]

According to the form recognition method of the present invention, it is possible to identify the type of a form with high accuracy even for an atypical form such as a registered notice.

According to the form recognition method of the present invention, an underline can be accurately extracted without being mistaken for a stroke in a character.

Further, according to the form recognition method of the present invention, the forms can be sorted based on the recognition result of the forms.

[Brief description of drawings]

FIG. 1 is a configuration diagram of a registration information recognition system that is an embodiment of the present invention.

FIG. 2 is a block diagram showing a process of recognizing registration information.

FIG. 3 is a PAD diagram of registration information recognition shown in FIG.

FIG. 4 is an explanatory diagram of an image of a registered notice that is a recognition target.

5 is a diagram showing a result of the ruled line extraction processing in step 304 of FIG. 3 performed on the image of FIG. 4;

6 is a diagram showing a result of performing the frame extraction processing of step 306 of FIG. 3 on the image of FIG. 4;

7 is a diagram showing the result of the character line extraction processing of step 310 of FIG. 3 on the image of FIG. 4;

8 is a PAD diagram regarding an underline extraction process of step 314 of FIG. 3. FIG.

FIG. 9 is an explanatory diagram of an underline extraction target image.

10 is a diagram showing a result of extracting ruled lines and character lines from the image of FIG.

FIG. 11 is a diagram showing a result of extracting ruled lines and cutting out characters from the image of FIG. 9;

FIG. 12 is an explanatory diagram of an image in which an underline is printed only on a part of characters in a character line.

13 is a PAD diagram related to the underline extraction processing in step 314 of FIG. 3. FIG.

FIG. 14 is an explanatory diagram of an underline extraction target image.

FIG. 15 is a diagram showing a result of extracting run-length data not forming a frame line and long underline from the image of FIG. 14;

16 is a PAD diagram relating to the underline extraction processing in step 314 of FIG. 3. FIG.

17 is a diagram showing a result of extracting run length data that does not form a frame line from the image of FIG. 14 and projecting the length of the run length data in a tilt direction.

FIG. 18 is a PAD diagram related to the form identification process in step 328 of FIG. 3.

FIG. 19 is a configuration diagram of a registration information recognition system having a sorting function, which is an embodiment of the present invention.

[Explanation of symbols]

200 ... Image input, 204 ... Ruled line extraction, 206 ... Frame extraction, 208 ... Table feature extraction, 246 ... Form identification, 222 ...
Character segmentation, 224 ... Character identification, 236 ... Word matching,
240 ... Content verification

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Kazuki Nakajima 1-280, Higashi Koikekubo, Kokubunji, Tokyo Inside the Central Research Laboratory, Hitachi, Ltd.

Claims (8)

[Claims] 1. A form recognition method for inputting a front side image of a form and reading characters, wherein character lines and frames are extracted from the front side image, and the name of the form is indicated from the positional relationship between the extracted plurality of character lines and frames. A first process for identifying the type of the form by selecting a line of text and reading the line of text indicating the name of the form, and extracting ruled lines from the surface image, and extracting the ruled lines from the extracted ruled lines to a table on the form. Second process for identifying the type of the form from the features of the table, extracting the character line from the top image, reading the extracted character line, and reading the extracted character line, the item of the form from the read result. A third process for selecting a name and identifying the type of the form from the combination of item names, and combining the processing results of the three processes to identify the type of the form. Form recognition method. 2. A registration information recognition method for inputting a surface image of a registered notice and reading characters, wherein a character line is extracted from the image of the registered notice and the positional relationship between the extracted plurality of character lines and a frame is extracted. The first method to identify the type of registered notice by selecting the character line of the form name and reading the character line of the form name, and the ruled line was extracted from the image of the registered notice and extracted. A second method of extracting the characteristics of the table from the ruled line and identifying the type of registered notification from the characteristics of the table, and identifying the type of registered notification by combining the results of the two methods. A form recognition method characterized by: 3. A registration information recognition method for inputting a surface image of a registered notice and reading characters, wherein a character line is extracted from the image of the registered notice and the positional relationship between the extracted plurality of character lines and the frame is extracted. The first method to identify the type of registered notice by selecting the character line of the form name and reading the character line of the form name, and extracting the character line from the image of the registered notice, and extracting The second method of reading the written character line, selecting the item name of the form from the read result, and identifying the type of registered notification from the combination of item names.
A form recognition method characterized by identifying the type of registered notification by combining the results of the two methods. 4. A registration information recognition method for inputting a front image of a registered notice and reading characters, wherein character lines and ruled lines are extracted from the image of the registered notice, and the extracted ruled lines are not frame ruled lines or frame ruled lines. The ruled lines are distinguished, the character lines in the frame that include the ruled lines that are not the frame ruled lines are detected, and the ruled lines that are not the frame ruled lines in the frame are determined based on the positional relationship between the character lines in the frame and the ruled lines that are not the frame ruled lines in the frame. A form recognition method characterized by determining whether is an underline. 5. A method of recognizing registration information in which a surface image of a registered notice is input and characters are read, by extracting an underline from the image of the registered notice and detecting black pixels existing on an extension of the underline. , A form recognition method characterized by determining the detected black pixel arrangement as an underline. 6. A method of recognizing registration information in which a surface image of a registered notice is input and characters are read, and a ruled line that does not form a frame ruled line is extracted from the registered notice image, and the value of the length of the ruled line is extracted. A form recognition method characterized by extracting an underline by projecting in the tilt direction and determining the projection value. 7. A method of recognizing registration information for inputting a surface image of a registered notice and reading characters, extracting a character line from an image of a registered notice, cutting out a character from the extracted character line, and cutting out the character. From the identified characters,
A form recognition method characterized by detecting characters corresponding to a location and a lot number and sorting registered notices in order of chome, street number, and issue for each town. 8. A method of recognizing registration information for inputting a surface image of a registered notice and reading characters, extracting a character line from an image of a registered notice, cutting out a character from the extracted character line, and cutting out the character. From the identified characters,
A form recognition method characterized by detecting characters corresponding to a creation date and a number and sorting registered notifications in order of creation date and number.