JPH06119491A - Slip comprehension system - Google Patents

Abstract

PURPOSE:To reduce labor at the time of preparing slip definition information and to improve the accuracy of recognition concerning the slip comprehension system for segmenting and recognizing characters described inside a description frame from a slip designating the description frame with a black frame line. CONSTITUTION:A slip definition table 108 is prepared for making physical arrangement relation between the items of the recognizing object slip correspondent to the class of descriptions such as an address and name or the like. A rectangle circumscrubing to the continuity of white picture elements is extracted from the image data of an input slip 101 by a description frame extracting means 105, and the rectangle provided with a size larger than a threshold value is decided as the description frame. A class judging means 107 judges the class of the described item by collating the physical arrangement relation of the items provided from the decided description frame with the definition information of the slip definition tavle 108. Further, the character string is recognized by a character string recognizing means 109 while using original knowledge corresponding to the class for each described item.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a form comprehension system for recognizing by cutting out characters entered in an entry frame from a form in which an entry frame is designated by a black frame line.

[0002]

2. Description of the Related Art FIG. 6 is a diagram for explaining a conventional technique. In FIG. 6, 601 represents an input form, and 602 represents a form definition table in which conventional form definition information is stored.

Conventionally, in order to detect the type of the character string entered in the entry frame, mapping is performed between the definition information regarding the type and the absolute coordinates of the entry position on the form (or the relative coordinates from the marker etc.). Was there. For example, as shown in FIG. 6, a form definition table 60 prepared in advance is used.
2, the x, y coordinates Sx, Sy at the upper left corner of item 1 and the sizes Lx, Ly in each direction of the entry frame are given, and the type (name in this example) entered at that position is also described. As a result, it can be seen that the character string written in the coordinates is a character string representing the name.

On the other hand, in the character string recognition, there are many methods for improving the recognition rate by using knowledge according to the type of the character string. For example, recognition using knowledge such that only words representing a person's name are used in the name field, and only numbers and symbols are used in the telephone number field.

[0005]

SUMMARY OF THE INVENTION In the above conventional method,
It is necessary to specify the coordinates of the entry frame in advance. For this reason, when the arrangement between the entry frames becomes dense, it is necessary to accurately specify the position coordinates, which is a factor of troublesomeness when creating the form definition information. In addition, even if the correct designation is made, it is not possible to deal with the misalignment error due to uneven printing of the form and the slight inclination at the time of inputting the form. When performing recognition, etc., the image cannot be cut out correctly, resulting in erroneous recognition. These phenomena are
Especially noticeable.

An object of the present invention is to provide a form understanding system which eliminates the need for precise specification of position coordinates, reduces the time and effort required when creating form definition information, and is resistant to misalignment and accurately recognizes each entered character string. To provide.

[0007]

[Means for Solving the Problems] In order to solve the above problems, a form definition table is prepared in which the physical layout relationship between the items of the recognition target form and the entry types such as the address and the family name are associated with each other. Every time, an image storage means for storing image data input from a scanner or the like, and a quadrilateral circumscribing a series of white pixels from the data are extracted, and a quadrilateral having a size equal to or larger than a threshold value is determined as an entry frame. The frame extraction means is collated with the physical arrangement relationship of the items obtained by the entry frame extraction means and the definition information of the form definition table,
The type of entry item is determined by the type determination means for determining the type of entry. Further, the image data is converted into code information and stored by the character string recognition means that performs character segmentation, character recognition, and word matching using unique knowledge according to the type for each entry.

[0008]

In the present invention, the entry frame is automatically detected from the quadrangle formed by the connected components of the white pixels, and the physical arrangement relationship between
Since the entry type is determined by collating with the physical arrangement relationship of the entry frames defined in the form definition table, it is not necessary to specify precise position coordinates as the form definition information. Also,
It is resistant to misalignment, and it is possible to recognize each written character string with high accuracy.

[0009]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing the configuration of an embodiment of the present invention,
FIG. 2 is a processing flowchart of the embodiment of the present invention, and FIG. 3 is a diagram showing a description example of the form definition table according to the embodiment of the present invention.

In FIG. 1, reference numeral 101 denotes an input form in which an input frame is designated by a black frame line, which is input by binary data read by an image input device 102 such as a scanner. Reference numeral 103 denotes a processing device for inputting a form, which includes a CPU and a memory. Image storage means 10
4 stores the image data input from the image input device 102 in the memory of the processing device 103 (process 201 in FIG. 2).

The entry frame extracting means 105 is provided for the input form 101.
The continuous white pixel area existing in the image data is detected, a circumscribing quadrangle is generated, and a frame is extracted (process 202 in FIG. 2). The processing of the entry frame extraction means 105 will be described later in detail with reference to FIG.

The input image is framed by black ruled lines. Therefore, in the case of an image with only the entry frame, the continuous white pixel area is only a closed area surrounded by the entry frame, and the number of entry frames can be made. However, in reality, since a character is entered in each item, there is a closed area surrounded by a character line. Therefore, the entry frame determination unit 106 in the entry frame extraction unit 105 determines whether the closed region of the entry frame or a part of the character is selected based on the size of the extracted rectangle, and selects only the closed region of the entry frame. (Process 203 in FIG. 2). The extraction process is repeated until the extraction of all the entry frames is completed (process 204 in FIG. 2).

The form definition table 108 stores a plurality of definition tables in which the types of respective items of the input form are described. FIG. 3 shows a description example of one definition table for the form definition table 108, in which the type of item, the type of item frame / character frame, the number of characters in the case of a character frame, etc. are described. Has been done. The character frame is a frame in which only one character is entered, and the item frame is a frame in which one or more characters are entered in a character string. In the example shown in FIG. 3, the first line is an address, the 12 character boxes from the left of the second line are telephone numbers, the remaining 5 character boxes are zip codes, and the third line is a name.
The fourth line defines the form for inputting the product name from the left, the next two character boxes are the model numbers, and the remaining two character boxes are the colors.

The type determination means 107 is the entry frame extraction means 1
The matching definition table is selected from the definition tables stored in the form definition table 108 by using the information such as the item layout and the number of consecutive character frames obtained in 05 (process 205 in FIG. 2). . In addition, this type determination means 1
The processing according to 07 will be described later in detail with reference to FIG.

The character string recognizing means 109 is a processing means for recognizing the character string image in each entry frame. The character string recognition unit 109 recognizes the character string using the knowledge obtained from the type of item obtained by the type determination unit 107 (process 206 in FIG. 2).

In this character string recognition, for example, the technique shown in the following reference can be applied. [References] Nakabayashi et al., "High-speed frame-free handwritten character string reading method using fuzzy term search", Theology, D-II, Vo
l.J74-D-II, No.11, pp.1528-1537.

First, character candidates to be recognized by the character dictionary 110 are extracted. Then, for an item of address, for example, a character string used in the address is prepared as a dictionary 111 (hereinafter referred to as a term dictionary), and character candidates are collated with the term dictionary 111. As a result, highly accurate character recognition can be realized.

FIG. 4 is a diagram for explaining an entry frame extracting means for extracting a continuous portion of white pixels. In FIG. 4, each circle is assumed to be a pixel, the outermost black pixel column indicates a black frame line, and the graphic existing in the center indicates the graphic entered in the item.

As shown in FIG. 4A, the entry frame extraction means 105 collects white pixels adjacent to each other in the up, down, left, or right direction from the upper end in the scanning direction as the same area (hereinafter, referred to as a book). The process is called labeling), and the coordinates of the four corners of the circumscribing rectangle are generated / changed. For example, up to line i, points A and B are labeled as points in different areas, but when processing line j, the circumscribing rectangle including point A and the circumscribing rectangle including point B are the white pixels of line j. They are connected by and are labeled as the same area. In this way, finally, a continuous region 401 of white pixels to be labeled as the same region is obtained, and each of the four rectangular regions is
It is possible to extract the absolute coordinates on the corner sheet.

FIG. 4B shows an example in which there are two continuous regions of white pixels. By the processing similar to the processing described in (a), continuous areas 402 and 403 are obtained and two rectangular areas are obtained.

FIG. 5 is an explanatory diagram of the type determining means 107 shown in FIG. In FIG. 5, reference numeral 501 indicates the physical arrangement of the entry frames extracted by the entry frame extraction means 105. Fifty
Reference numeral 2 is entry frame information obtained by analyzing the layout 501, which corresponds to the layout of the entry frame to create an item frame and a character frame.
It shows the internal state of the memory expanded into a two-dimensional array. Reference numerals 503 and 504 denote entry frame information obtained from the form definition table 108.

In creating the entry frame information 502, each entry frame is arranged in rows and columns according to the center coordinates of the entry frame. Also, each entry frame is classified into an item frame and a character frame according to the size of the frame, and regarding the character frame, the consecutive character frames are collected and the number is obtained.

On the other hand, the entry frame information 503, 504 of the form definition table 108 is a physical layout relationship defined in advance, which is expanded into a two-dimensional array like the entry frame information 502. As shown in FIG. 3, the form definition table 108 defines the entry type, the character frame / item frame, and the number of characters in the case of a character frame.
The entry type is added to 02.

The type determining means 107 collates the input form with a plurality of definition tables by using the item frame / character frame distinction and the number of character frames as keys, finds a matching definition table, and finds each of the input frames. Ask for the type of entry. For example,
The entry frame information 502 corresponding to the physical arrangement 501 of the entry frames extracted by the entry frame extraction means 105 has an item frame at the first row and the first column, and an entry frame at the first row and the second column. Since it does not exist, it can be seen that it does not match the definition of the entry frame information 504 in the definition table.

With respect to the character frames, a process of once combining all the adjacent frames for each line (for example, in the example of FIG. 5, all the adjacent character frames are combined into one in the third line). If the character frames are adjacent to each other on the definition table side, the number of characters is added and then the comparison is performed. In this way, the entry frame information 502 matches the entry frame information 503, and the entry type is given to each frame. The character frame is divided according to the definition table at the time of recognition.

When a document is input with an inclination, the entry frame, which is a continuous area of white pixels, is extracted with an inclination. Therefore, the inclination of the horizontal line can be calculated for any of the entry boxes. The center coordinates of each entry frame are corrected based on this inclination information, and the entry frames in the same line are detected.

[0027]

As described above, according to the present invention,
Since the entry frame written with black ruled lines is detected by the continuous white pixel portion, detailed definition such as the absolute coordinates of the entry frame becomes unnecessary. Also, when extracting a continuous area of white pixels, it is possible to detect an inclined entry frame by obtaining the coordinates of the four corner points of the circumscribed rectangle. As a result, the shift due to the inclination of the image can be processed by the same means.

Further, since the entry type of the input form can be determined by using the definition table in which the physical layout relationship of the entry frame and the entry type are associated with each other, unique knowledge for each entry type is used at the time of character recognition. The recognition accuracy can be improved.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of an exemplary embodiment of the present invention.

FIG. 2 is a processing flowchart of an embodiment of the present invention.

FIG. 3 is a diagram showing a description example of a form definition table according to the embodiment of the present invention.

FIG. 4 is an explanatory diagram of an entry frame extracting unit according to the exemplary embodiment of the present invention.

FIG. 5 is an explanatory diagram of a type determination unit according to the exemplary embodiment of the present invention.

FIG. 6 is a diagram for explaining a conventional technique.

[Explanation of symbols]

 101 input form 102 image input device 103 processing device 104 image storage means 105 entry frame extraction means 106 entry frame determination means 107 type determination means 108 form definition table 109 character string recognition means 110 character dictionary 111 term dictionary

Claims (1)

[Claims] 1. From a form image in which at least one entry frame surrounded by a black frame line exists, a character string entered in each entry frame is entered in the entry type determined in advance according to the entry frame of the form. In a form understanding system for classifying and recognizing, an image storing means (104) for storing image data of an input form (101) inputted from an image input device (102) and a quadrilateral four points formed by connected components of pixels of the same color. Corresponds the entry type and the position of the entry frame with the entry frame extraction means (105) for extracting the vertical and horizontal coordinates of each of the above and selecting the quadrangle by selecting the quadrangle by comparing it with a predetermined threshold value. The form definition table (108) for storing the selected information and the form definition table (10
Type determination means for determining the type of entry frame by collating the physical placement relationship between entry boxes defined in 8) with the physical placement relationship between entry frames obtained by the entry frame extraction means (105) (107) and a character string recognition means (109) for performing character cutting, recognition, and knowledge processing on the character strings in the entry frame by using knowledge specific to the entry type. Understanding system.