JPH02230366A - Slip recognizing device - Google Patents

Abstract

PURPOSE:To recognize a slip for which the pre-print of an ID number, etc., is hard to preform, by identifying a slip format by the difference of feature of a table field included in the slip generally. CONSTITUTION:A slip input part 11 converts a character and a mark, etc., recorded on the slip to binary image data in black and white, and a white connection rectangular extraction part 12 extracts the connection part of a white picture element in a slip image, and furthermore, outputs the position coordinate of a rectangle(white connection rectangle) in parallel with each side of an input slip circumscribing with each connection part. Also, a feature vector calculation part 14 outputs a feature vector representing a feature with respect to the position and the size of the white connection rectangle included in the input slip. And an identification part 17 selects the minimum distance out of distance outputted from a distance calculation part 15, and recognizes the character in the input slip according to the format obtained at a slip recognition part 18. Thereby, the slip format can be identified without pre-printing the ID number, etc., for the identification of the slip format.

Description

DETAILED DESCRIPTION OF THE INVENTION [Industrial Application Fields] The present invention relates to an image recognition processing device, in particular, an image recognition processing device that has a memory for storing a plurality of format information, identifies the format of an input form, and uses the format information to identify the format of an input form. The present invention relates to a device that recognizes forms based on the following information.

[Background Art] With advances in computer-based character recognition technology, optical character readers (OCR) have been developed to automate the work of inputting data into computers in offices and the like. OCR
To input data, write characters on a form (OCR form) in a format registered in advance in OCR, and OCR inputs this form as an image, cuts out the characters according to the format information stored in memory, and recognizes them. Realized.

On the other hand, since OCR forms have different formats indicating the position of characters to be read and the processing method of recognition results depending on the type thereof, it is necessary to identify them prior to recognition processing.

Therefore, when reading multiple forms with different formats using OCR, either manually specify the format of the form to be input from among the registered format information, or set the format in a fixed position common to all registered form formats. A method is used in which an ID number or the like for identifying the type is pre-printed and this is recognized prior to recognizing the form.

[Problem to be solved by the invention] Conventional form recognition devices such as OCR pre-print an ID number for identification at a fixed position on the form to be input, and recognize this before recognizing the form. The document format is identified using methods such as However, when attempting to recognize a wide variety of existing forms, it is difficult to uniformly request pre-printing of ID numbers, etc. at specific positions on these forms, and may not always be possible. However, as long as prior art is relied upon, there is a problem in that these forms cannot be recognized without pre-printing ID numbers and the like.

An object of the present invention is to recognize existing forms, such as existing OCR forms, in which it is difficult to pre-print an ID number or the like to identify the format type.

[Means for Solving the Problems] The present invention uses differences in characteristics regarding table fields generally included in a form to print a form without pre-printing an ID number or the like for identifying such a format on the form. It is characterized by identifying the format, and has the following means for this purpose.

(a) Form input means.

(b) White connected rectangle extraction means.

(c) Feature vector calculation means (d) Distance calculation means (e) Form recognition means [Effect] The device of the invention works as follows.

(a) The form input means writes a black and white binary image into the image memory by scanning the form.

(b) The white connected rectangle extraction means focuses on the fact that the table field in which character data is written in the form is a white rectangular area surrounded by ruled lines parallel to each side of the form, and extracts this. Connected components of white pixels of the form image written in the image memory are extracted, and the position coordinates of a rectangle (hereinafter referred to as a white connected rectangle) that is circumscribed to each connected component and parallel to each side of the input form are output.

(c) The feature vector calculation means uses the results obtained by the white connected rectangle extraction means to calculate a feature vector representing characteristics regarding the number, position, and size of white connected rectangles included in the input form. Here, if the input form is a learning form, the feature vector is stored in the format memory.

(d) When the input form is a test form, the distance calculation means:
The feature vectors for each type of form stored in the format memory are read out, and the distances between these and the feature vectors obtained by the feature vector calculation means are determined.

(e) The form recognition means selects the closest one among the outputs obtained by the distance calculation means, reads the format from the format memory, and recognizes the characters in the input form according to the format.

Embodiment] FIG. 1 is a block diagram showing the configuration of an embodiment of a form recognition device of the present invention. The form input unit 11 scans the form, converts characters and symbols recorded on the form into black and white binary image data, and writes the data into the image memory unit 13.

The white connected rectangle extraction unit 12 extracts this by focusing on the fact that the table field in which character data is written in the form is a white rectangular area surrounded by ruled lines parallel to each side of the form. Connected components of white pixels of the form image written in the section are extracted, and the position coordinates of a rectangle (hereinafter referred to as a white connected rectangle) that is circumscribed to each connected component and parallel to each side of the input form are output. However, since some white connected rectangles are extracted from closed areas in characters in addition to table fields, a threshold value is used to remove these small white connected rectangles.

FIG. 2 shows an example of an input form, and FIG. 3 shows a white connected rectangle outputted from the white connected rectangle extraction unit l2 using the input form of FIG.

The feature vector calculation unit 14 shown in FIG. 1 uses the output of the white connected rectangle extraction unit 12 to output a feature vector representing features regarding the number, position, and size of white connected rectangles included in the input form. Details of the feature vector output means will be described below.

Information about where each connected rectangle (for example, the white connected rectangle 31 in FIG. 3 extracted corresponding to the table field 2l in FIG. 2 in FIG. 3) included in the input form is located in the top, bottom, left, and right of the form. In order to output it as a feature vector, the input form image is divided vertically and horizontally into M equal strips. Here, the i-th (i=1,...,
The strip area of M) is labeled X, and the j-th ('=1...,M) strip area from the top, which is divided into M equal parts in the horizontal direction, is labeled Y.

Next, in order to output information about how large the white connected rectangles are included in the input form as a feature vector, we calculate the horizontal and vertical lengths of each connected rectangle to the horizontal and vertical lengths of the input form, respectively. , and quantize this into N stages,
The length (height) of each connected rectangle in the vertical direction is L(k=1,・
・・・． N), the horizontal length (width) is L(]=1,...
．． Label it N).

The feature vector is, for each strip area X and Y,
It is calculated by counting the number of white connected rectangles with sizes Hk and L, respectively.

The following three methods are possible for counting.

■ Counting in the area that includes the center of the white connected rectangle, ■ Counting in the area that includes at least part of the white connected rectangle, ■ Weighting is performed according to the ratio of the white connected rectangle to the size of the area. count.

Here, the number of dimensions D of the feature vector is D=2 x (Mx (N x N)). Fourth
The figure shows an example of a feature vector for the case of M=N=2.

Here, when the input form is a learning form, the feature vector calculation unit 14 in FIG. .

When the input form is a test form, the distance calculation unit 15 in FIG. Calculate the Euclidean distance and print the result.

The identification section l7 in FIG. The characters in the input form are recognized according to the specified format.

Figure 5 shows how to count white connected rectangles for four types of forms A, B%C, and D (10 forms are used for each type, 7 are learning forms and the remaining 3 are test forms). Using the method of ■, find the distance between the average feature vectors of the learning samples when M = N = 2, and the distance from the average feature vector to the farthest sample in each form category, and calculate the distance occupied by each category space. The range is shown by a solid line.

The right side of FIG. 5 shows the category spaces 51, 52, 53, and 54 of four types of forms ASBSC and D, and it can be seen that there is no overlap between them and that they can be correctly identified.

Here, if the number of divisions M and the number of quantization N are made smaller (for example, M=N=1), the number of dimensions of the feature vector becomes smaller and the amount of calculation is reduced, but the position and size of the white connected rectangles are Therefore, if a form is input at an angle, if a closed area in text is mistakenly extracted as a table area, or if a table area is divided due to noise, etc., the identification ability will deteriorate. Since it is expected that M
=N=2 or more is desirable.

[Effect of the invention] As described above, according to the present invention, the conventional OCR
In a form recognition device typified by , it is possible to identify a form format without pre-printing an ID number or the like for identifying the form format.

Therefore, in addition to conventional OCR forms, it is possible to recognize a wide variety of existing forms, and the types of forms on which data can be input manually are significantly expanded.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing an example of the configuration of a form identification device, FIG. 2 is a diagram showing an example of an input form, FIG. 3 is a diagram showing an example of white connected rectangles, and FIG. 4 is a feature vector A diagram showing an example of
FIG. 5 is a diagram showing an example of identification ability for four types of forms. : Form input section: White connected rectangle section: Image memory section: Feature vector calculation section: Distance calculation section: Format memory section: Identification section: Form recognition section Two-table field: White connected rectangle: Category space of form A: Category space of form B Category space: Category space of form C2 Category space of form D Applicant NTT Data Communications Co., Ltd. Agent Patent attorney Takashi Kanehira Figure 1 Figure C Figure 5 Procedural amendment (method) 1999 June 22nd

Claims (4)

[Claims] (1) In a form recognition device having a memory for storing a plurality of format information, means for inputting a form as a black and white binary image, means for determining a rectangular frame circumscribing a connected component of white pixels in the input image, and Average characteristics vector obtained from the learning sample for the method of calculating the number of table fields included in the input image and the characteristics vector that represents the number, position, and size, and this feature vector and the format forms of multiple different formats. A form recognition device comprising: means for determining a distance; and means for processing an input form according to format information of a form having the closest distance. (2) In the apparatus according to claim 1, the means for calculating the feature vector includes dividing the input form image into small areas and counting the number of table fields in the area including the center of the table field. Characteristic form recognition device. (3) In the apparatus according to claim 1, the means for calculating the feature vector is to divide an input form image into small areas and count the number of table fields in the area including at least a part of the table fields. A form recognition device featuring: (4) In the apparatus according to claim 1, the means for calculating the feature vector divides the input form image into small areas, and in the area including even a part of the table field, the whiteness relative to the size of the area is A form recognition device that counts the number of table fields by weighting according to the proportion of the size occupied by connected rectangles.