JPS60146373A - Pattern recognition system - Google Patents

Abstract

PURPOSE:To facilitate specification of format information by utilizing a document form where characters (including marks) are entered and by feeding out and recognizing them. CONSTITUTION:A control part 2 feeds-out a character pattern from a document form image. An output of the feed-out processing includes a character pattern and its position information, which are stored in a memory 5. A character pattern is recognized in a character recognition part 6, and a character code of the recognized result is stored in the memory 5 as related to the position information. Characters which cannot be recognized are displayed on a keyboard display 7, and an operator inputs a right answer. After all characters have been recognized, the control part 2 executes field generation processing. An input document form is converted into a digital image in a photoelectric converter 4 and stored in the memory 5. The character recognition part 6 carries out character feeding-out and recognition through a known method by utilizing format information in terms of the documents form image stored in the memory 5.

Description

[Detailed Description of the Invention] [Field of the Invention] The present invention relates to a butter 7 recognition method, and is particularly a character/symbol recognition method suitable for controlling an output format based on information on the position of writing characters on a form. Regarding.

[Background of the invention]

Conventional character recognition devices define a character frame based on the position measured from the edge of the form, and specify an output format (for example, column position on a card image or check format) for that character frame.
Screen style information that instructs the method of checking calculations, etc.) was specified for each form type.

Hereinafter, information stored along with a form, such as the position of a character frame, output format, and check format, will be referred to as format information. However, since the method of specifying format information is difficult, there is a need for a method that simplifies this process.

In addition, existing forms do not have character recognition in mind, so some do not have character frames, and it is difficult to specify character frames for such documents. Furthermore, black characters, tables, lines, etc. (hereinafter referred to as blemish prints) may act as interfering information, and it is often difficult to cut out the characters to be read.

[Purpose of the invention]

SUMMARY OF THE INVENTION In view of the above-mentioned drawbacks, an object of the present invention is to provide a pattern recognition device that can easily specify format information. An object of the present invention is to provide a pattern recognition method that allows format information to be easily specified.

[Summary of the invention]

In order to achieve the above object, in the present invention, characters (including symbols) are written on a form, and these letters are cut out and recognized. Since the position of the characters can be measured when cutting out the characters, the position information of the character frame can be generated from this, the character recognition result can be understood to generate information such as the output format, and from this, a set of i% The above purpose is achieved by generating [.

[Embodiments of the invention]

Hereinafter, a one-dog embodiment of the present invention will be explained with reference to FIG. 1st
The figure is a block diagram showing the configuration of a device implementing the recognition method of the present invention. Each part of the device is connected to a bus 1, and the overall operation is controlled by a controller 2. - The information on ticket 3 is stored in the memory 5 through photo-transformation + tt4 photo-transformation, phototransformation, and digitization. Although not described in the following, saving the memory required for expressing the image by efficient encoding may also be advantageous in terms of cost. Further, in the following explanation, it is assumed that one pixel is binarized into one bit, but one pixel may be expressed with multiple values, or the color information may be expressed by a color scanner.

When using the form, first write characters indicating format information on the form to be used (hereinafter referred to as the full format instruction form).
have them read. In this embodiment, it is assumed that the form is printed in a dropout color, and that only the characters written by Kijin Bunte are expressed in black in the form image. As for the characters used here, only uppercase alphabetic letters are used for ``k'' to simplify the extraction process described later, but not only alphabetical characters but also katakana, numbers, kanji, etc. may be used.

A character pattern is cut out for 1flJ control section 2 from the above 1-slip image. Details of the extraction process will be described later. The output of the cutting process is a character pattern and its position information 69, which are stored in the memory 5. The character pattern is recognized by the sentence recognition section 6, and the character code resulting from the recognition is stored in the memory 5 in a form that is associated with the position information. When an unrecognizable sentence is rarely found, it is displayed on the keyboard display 7, and the operator inputs the correct answer. After all character patterns have been recognized, the control section 2 performs field generation processing. In this embodiment, for example, for a character written as shown in FIG. 2 (A), a circumscribed quadrilateral of a letter is created as shown in FIG. 2 (l), and a circumscribed quadrilateral for the same character is A rectangular area as shown in Figure (C) is generated. This area corresponds to what is called a field in the sentence recognition device, and represents a frame in which a series of sentence data is written. Details of the field generation process will be described later.A different Fumiko code is used for the valley field.)
This character code is used as identification information for the field, and by counting the number of characters in the field, the number of characters entered in the jt dog can be determined. The process of associating field identification information with various format 1H reports is carried out by the control unit 2. The correspondence will be described later in the processing, and the # measurement will be described later. The above-mentioned maximum number of characters to be entered is also included in the one-format information.

The format information is stored in the memory 5, while the output device t8
You can also output to It is also effective to display the entire screen-style instruction information as a character image on the keyboard display 7.

When the processing of the i-type instruction form is completed, processing of the input form begins. The manual form is converted into a digital image by the photoelectric conversion device 4, and the form image is stored in the memory 5. Stored in the memory 5 For the above form image, the character recognition section 6 extracts and recognizes the sentence using - Take 1# (in the memory 5) according to the official method. I do. Although the recognition result is also checked using the screen type 1# report, either the check processing card recognition unit 6 or the control product 2 may perform the check. The results are transferred to the output device, and the process moves on to the next input slip.

The details of the processing in the 111 control unit 2 will be described shortly. Figure 3 shows the flow factors of the extraction process.
gram Analysis+s l)iagram) format. 301 is initialization and sets the number of contours K to 0.
Set to . A double loop of steps 302 and 303 is used to scan one square image P (iIJ), and the processes of steps 304 to 308 are repeated. In 304, black pixels are constructed. Hidden/
Kotoki, 305-310 place 4! I do. K in 305
Increase by 1. In the loop of 306, the outline is traced using the black pixel created in 304 as a starting point, and in 306, the minimum value and maximum value of the x+yffl mark in that area are traced.
a HXmmx +)'1111111)'5axk
lfl out. In 307 (Xm1m gyall and (
A quadrilateral structure having diagonal vertices at XYALL YTA & ENG is transmitted to the first area of the pattern buffer (area in the memory 5). In 309, delete the ``direction'' of the above quadrilateral ' song area in the form image A (P(x,
Set y)=0 or 2 and perform process 7. At 31υ, record the center coordinates (, xh, Yk) of the above quadrilateral castle. It is clear that the above-described processing cuts out the Bunko patterns, and K represents the number of Bunko patterns at the time when the J-print image is scanned.

FIG. 4 shows a flowchart of field generation processing in PAD format. 400, the initial value of the number of fields M is 0.

401 is a loop that repeats 402-413 as many times as Fumiko. 402, the character recognition result 几(k)
Check whether is a space. If it is not a space, perform steps 403 to 413. In step 403, initialization is performed such as incrementing the field number M by 1 and decrementing the initial value of the number of characters LM in the field by 1 and 2. 404 is a loop in which it is checked whether t is the same as 几(k) for the recognition result of k+1 or more; red. According to the determination in step 405, when the jth character is the same as the first character (R(j)=R(k)), steps 406 to 408 are performed. 406 is the number of characters, 4'j0'4 Ll 7 is the center position of 6 sentences (xJlyJ's'), and the small value x,
, y, and the maximum value xb.

'jb and the director. In 408, kL(N is replaced with a space so that the same process will not be performed again. When the loop in 404 is completed, the number of times of the character with the result, and the maximum and minimum values of its coordinates. 409 compares the difference between the maximum and minimum values for He swears in one line.This honor is not a necessary one, but is a temporary one set to simplify the explanation.

) In the case of a vertical screen, the process of 410 is performed, and the upper and lower δ of the average value of the X coordinate are set as the horizontal duck of the field.

The vertical length is determined by leaving a margin of δ outside between the minimum and maximum values. When writing horizontally, process 411 is performed.

412 is the transfer of field information, field number M1 character e L M, t% mark (X
's, Xb, Y-, Yb), field identification information R(k) is transferred. At 413, clear with R(k) meeting space. Therefore, when the loop 401 is completed, the information of the upper i4 is transmitted for the field 2, and all of the fields 2 and 4 become spaces. Note that non-processing operates correctly even when the number of text children in the field Lw is 1, which is 4.

In addition, the pitch of Fumiko's entry may be calculated from the number of characters in field 1--E and sent.

Next, FIG. 5 shows a processing diagram of the correspondence between the field identification 1# information and the oath ceremony information in PAD format.

In this process, the confirmation of identification 1 is strictly ordered.

501 is a loop that repeats the following processing by the number N of pieces of weight type information registered in advance. The n-th Sai-style Newsletter includes the character code G(n) as heading information. At 502, R and Q(n) are compared. If they match, in step 503, the information is transferred to the livestock type information area for the 04th d-hole information r field.

In the example, the explanation was given using English letters as the letters to be filled in on the honor style instruction form, but as mentioned above, it is not limited to these, and kana and kanji can also be used. It can be written in just a few lines, or it can be written in an L shape or a single character shape. 'Furthermore, as field specifications, it is possible not only to specify a quartet of the same sentence, but also to use various abbreviations and to use a frame as an auxiliary symbol.

FIG. 6 shows examples of various field specification modifications.

In Figure 8, (A) is a double character using Naoto as the connecting name. In (H), the area surrounded by a frame is one field,
This is an example in which the field identification symbol (represented by the letter A) and the same number of characters in the field (5 in this example) are Japanese and Japanese. (
In C), only the field identification symbol is written in the frame, and the number of characters in the same field and the character pitch are obtained from the Homare style information corresponding to the identification symbol. When this field identification symbol is a specific character, this field can be set to an image area so that only the image is transferred without cutting out or recognizing the text. (JJ) is an example in which the top designation is made only in the four corners, and (E) is an example in which the IA@ designation is made in square brackets as l/Ll.

The line 〇 extraction necessary for the above transformation will be explained.

FIG. 7 is a flowchart showing an example of processing for the rectangular structure shown in FIG. 7
Repeat steps 03 to 707 nine times and check whether the outline is rectangular or not. 703 resets error 7 lag. 70
In loop 4, the process goes around the contour and repeats the four corner extraction process in 705. In loop 706, the process goes around the contour and calculates the distance between each point and the virtual four sides calculated from the four corners.

If this distance is greater than or equal to the required distance ε, an error flag is set in C1707 where the virtual side and the contour are far apart (not directly aligned), and the loop is exited. If the error flag is in a reset state when the loop 706 is completed, the inner contour is assumed to be a rectangle. Although I have omitted the explanation, it is easy for young children to find out that the angles formed by two adjacent imaginary sides are almost right angles.

The second implementation/column will be briefly explained with reference to Figure 1#48. This embodiment has the same four components as in FIG.
The flow of b1JiII41 is different. The non-example 1 slip used is 9 with a frame and letters printed in the common fM color.This frame and Bunkokai are referred to as preprints hereafter. Figure 8 is a flow diagram (PAD
format). 801r1. Performs various initializations. 80
In step 2, V is stored in the memory 5 as an image of a form with letters and symbols written on it (an image of a blank form). Next, in 803, honor style instruction - form image B' of the invoice
Insert 9 into the C mesori.

At 804, the blank form image A is subtracted from the honor ceremony instruction form image B to obtain a single form image P that includes only the character pattern filled in for the oath ceremony instruction. Reference numeral 805 is the process of generating sentence cutout, 6g iron, and format information in the first embodiment, and oath ceremony information F can be generated. 806 and subsequent steps are the input form processing loop that should be cleared, 806
indicates that the processing will continue until there are no more input forms. 8
In step 07, the human power/expansion image C is loaded into the memory.

In 808, image C7Jh and image A are all subtracted,
A form image Q is obtained that criticizes the target Fumiko pattern. In 809, characters are cut out from the above image Q〃λ using d formula creation 4i F'x, and the characters are cut out into 7 squares.

810 outputs g 砿Karashi. When there are no more votes left, the process exits from the loop at 806 and completes the process at 811.

In the first and second central examples, it was assumed that the manual form [optical conversion] would be processed immediately, but any of the blank form, 41: formula instruction form, and input I area form could be processed by Rurui or When all of them are photoelectrically converted, the form image r may be stored directly in an external storage (for example, an optical disk), and then the form image area protruding from the external storage device may be processed. In this way, a document that is initially stored in the form of an image on an optical disk can be processed at a moment's leisure, and only a portion of it can be recognized and encoded. It becomes possible to encode only the variable parts in oaths, estimates, and other fixed forms.

〔Effect of the invention〕

According to the present invention, it is possible to realize a character recognition device that is extremely easy to use because it uses format information in a very simple form. Roughly speaking, it is possible to express the honor style instruction information on the form in a simple form, and then define detailed information separately and connect it, so it is necessary to add and modify the format information. Furthermore, existing forms can be easily used for sentence recognition, which saves the effort of designing new forms and printing them.

[Brief explanation of the drawing]

Fig. 1 is a block diagram showing the four components of one embodiment of the present invention, Fig. 2 is a principle diagram for explaining the form image, Figs. A flowchart of the process of opening is displayed. 1...Bus, 2...+ttlJ 4th part, 4...Photoelectric conversion part, 5...Me 2nd day (A) (B) (SuJ\, 3rd eye 4th mouth [Iku. − No. 5 (D) (E) No. 7 Concave No. δ Continuation of the first page @ Inventor Iken Masao Elementary School I Kaoru

Claims (1)

[Claims] 1. Recognize the symbols written on the input form, determine the relationship between the recognized symbols, and detect the area where symbols belonging to the same group exist, and the detected area A pattern recognition method characterized by generating format information from information and symbols. 2. In the description of claim 1, a set of the same symbols is identified among the recognized symbols and determined to belong to the same group, and a conversion table between the above symbols and the format 1# information is created. A pattern recognition method characterized by being provided separately. 3. In the description of claim 1, a line is extracted, an area surrounded by the line is detected, and a relationship between a recognized symbol and the area surrounded by the line is determined, thereby identifying symbols belonging to the same group. A pattern recognition method characterized by detecting existing areas. 4. In the statement of claim 1, after storing a standard form image converted into a digital image in advance and erasing the portion included in the standard form image from the input form image, , a pattern recognition method characterized by symbol recognition. 5. In the description of claim 1, is there an image of input 11? A pattern recognition method characterized by recognizing an input one-square image read out from the external storage device after it is once stored in the external storage device.