JPH01292586A - Back-up device for recognition of character - Google Patents

Abstract

PURPOSE:To ensure the sure correction of the misrecognized characters by using a pattern matching part to secure the matching between the character pattern included in a picture and the character pattern corresponding to a character code. CONSTITUTION:The image data read out of a picture reading part 101 is evolved to a bit image memory 103. A character pattern is segmented out of the memory 103 by a character segmenting circuit 104 and a character recognizing part 105 recognizes the character pattern to be recognized. Then a character generator 107 produces a character pattern corresponding to a recognized character code. A pattern matching circuit 115 secures the matching between the character pattern produced by the circuit 107 and the character pattern to be recognized. Then a deciding means of the circuit 115 decides whether the character code obtained at the part 105 is misrecognized or not. If so, a fact that the corresponding character pattern is misrecognized is informed to the outside. Thus the misrecognized character pattern is surely corrected.

Description

DETAILED DESCRIPTION OF THE INVENTION [Industrial Application Field] The present invention relates to a character recognition support device, particularly a character recognition support device that supports processing for misidentified expressions after character recognition.

[Prior Art] In general, it is practically impossible for character recognition devices to obtain 100% perfect recognition results, and the recognition process itself may be rejected or erroneous recognition may occur during the recognition process. is unavoidable.

Therefore, in order to obtain a document that completely matches the original to be recognized, it is necessary to correct the rejected or erroneously recognized characters by some means.

By the way, recognition rejection means that the character to be recognized is unrecognizable, and a specific character code is usually assigned to that character. Examples of this specific code include, for example, one that corresponds to a solid black character pattern, etc.
- This allows the operator to know by looking at the characters that the characters have been rejected.

On the other hand, misrecognition is when a character different from the original character is output as a recognition result.

[Problem to be Solved by the Invention] Therefore, when displaying the recognition result on the display screen and performing correction processing, as explained above, for rejected characters,
Since a special code is assigned as a penalty, a mark that stands out compared to normal characters will be displayed, and the correction operation will proceed relatively smoothly. This has to be done while checking the text, making it extremely troublesome work. Moreover,
Naturally, when this work is performed over many pages, oversights may occur, which remains a problem.

By the way, it is conceivable to install a natural language processing means as a post-processing of the recognition process and automatically correct incorrect characters so that they become correct words or sentences, but providing such a function would require a great deal of cost. In addition, considering the current situation where the natural language processing system itself has not been established, it cannot be said to be a good idea.

For example, language processing is not applicable to strings of numbers or characters that do not make sense as words or sentences (such as °'777-111Bd"), so in any case, humans will ultimately have to confirm or correct the recognition results. I have no choice but to do it.

The present invention has been made in view of this problem, and it is an object of the present invention to provide a character recognition support device that allows erroneously recognized characters to be read at a glance.

[Means for Solving the Problem] In order to solve this problem, the character recognition support device of the present invention has the configuration shown below.

That is, a recognition means for recognizing a character pattern to be recognized, a pattern generation means for generating a character pattern corresponding to the character code obtained by recognition, and a matching between the generated character pattern and the character pattern to be recognized. a matching means; a determining means for determining whether or not the character code obtained by the recognizing means is due to misrecognition based on the result of the matching means; and notifying means for notifying the outside that the character pattern to be recognized has been erroneously recognized.

[Operation] In the configuration of the present invention, the pattern generation means generates a character pattern corresponding to the character code obtained by the recognition means, and the matching means matches them. If the determining means determines that there is an erroneous recognition, the notifying means issues a notification.

[Embodiments] Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings.

<Explanation of the outline of the configuration (FIG. 1)> FIG. 1 is a block configuration diagram of the character recognition support device in this embodiment.

In the figure, 101 is an image reading unit that reads an original image, and 102
is an input unit such as a keyboard or a pointing device, which are electrically connected to the system bus 113 via the I10 ink face 112. Also,
103 is a bit image memory that expands the image data read from the image reading unit 101; 104 is a character cutting circuit that cuts out the character pattern in the bit image memory 103; and 105 is a character recognition circuit that recognizes the cut out character pattern. 106 is character cutting circuit 1
This is a character information memory that stores information related to characters based on information from 04 and the character recognition circuit 105. A character generator 107 generates a character pattern corresponding to a character code. Reference numeral 110 indicates a character code memory for storing a character code, which is the final purpose, and 116 indicates a VRAM for developing an image to be displayed. In this embodiment, it is assumed to have the same size (capacity) as the bit image memory 103 described above. . 117 is a graphic drawing circuit that draws lines, rectangles, etc. in the VRAM 116; A display 114 displays image data developed in the VRAM 116. 115 is a pattern matching unit, the details of which will be described later, matches the recognized optical character pattern with the character pattern corresponding to the character code obtained as a result of recognition, and processes whether or not they match. It is something that does.

111 is a CPU that comprehensively controls the entire device, and contains a CPU 11 processing procedure (program).
The computer is equipped with a ROM 111a that stores , and a RAM 111b that is used as a work area. The ROM 111a stores programs related to correction processing based on recognition results as well as programs related to the flowchart shown in FIG. 2.

<Explanation of Processing Contents (FIGS. 2 to 5) The specific processing contents of the apparatus of this embodiment having the above-described configuration will be explained using FIGS. 3 to 5 and according to the flowchart in FIG. 2.

First, in step S201, each memory and register are initialized, and in the next step 5202, it is determined whether there is an instruction from the input unit 102 to end a series of character recognition processing. Then, the processing from step 5203 shown below will be repeated until a termination instruction is issued.

In step 5203, one page of the document to be character recognized is set, the image reading unit 101 is driven and read, and the binary image data is transferred to the bit image memory 103.
Expand to. Next, in step 5204, the character cutting circuit 104 is activated to detect the size and position of each character pattern in the document image stored in the bit image memory 103. The size and position information of the detected character patterns are sequentially stored in the character information memory 1'06.

To give a specific example, assume that a character pattern 301 as shown in FIG. 3 is developed in the bit image memory 103. Note that 302 is a circumscribing rectangle of the character pattern 301.

Now, each pixel of the bit image memory 103 is (x, y
), and the pixel position at the upper left corner thereof is set as the origin (0,0). Furthermore, the circumscribed rectangle 302 has the coordinates (x+, yl) of its upper left end and the size W 1 + h
X l+ 'J l+ of each detected character, represented by +
W l+h l is stored in the character information memory 106.

FIG. 4 shows the contents of the character information memory 106. Circumscribed rectangle data of each character obtained by the character cutting circuit 104 described above X l+ y 1. W I+ h + is character position part 4 at character number i in character information memory 106
It is stored in 01. Character numbers are assigned in the order of appearance of the characters in the document, such as 1, 2, 3, and so on.

Now, in step 5205, the character information memory 106
xl+yl, W l+hl in the character position part 401
With reference to , each character pattern is sequentially transferred from the bit image memory 103 to the character recognition circuit 105 via the bus 113 to perform character recognition processing. Then, the character code corresponding to each character number i obtained as a recognition result is stored in the candidate character section 402 of the character information memory 106. - When fishing on a boat, it is possible for a character recognition device to output multiple character codes for one pattern in descending order of recognition certainty. , CI□... In addition, the number of candidate characters depends on the ability of the recognition circuit, but 2
The candidate characters after the number are used to correct the error in the first candidate character CIO. Incidentally, since various methods have been proposed for cutting out character patterns and character recognition itself in steps 3203 to 5205 described above, the details thereof will be omitted here. In addition, in the embodiment, since it is sufficient to perform character recognition processing that obtains the circumscribed rectangle of each character pattern and the character code for that pattern, the character extraction unit 104 and the recognition unit 105 that generates the character code are not necessarily clearly separated. There's no need to be.

Now, with the above processing, the bit image memory 10
After cutting out and recognizing each character pattern expanded in 3, the process proceeds to step 5209, where the character pattern to be recognized is matched with the character pattern corresponding to the recognized character code. .

This determines whether the character pattern to be recognized matches the character pattern corresponding to the character code obtained through recognition processing, that is, whether the character code obtained by recognition is due to misrecognition. That's what I'm trying to judge.

The specific processing contents of the pattern matching processing will be explained below.

FIG. 5 shows a pattern matching circuit 115 in the embodiment.
It is a figure showing the internal structure of.

CPUI 11 is character number i in character information memory 106
reads the character patterns indicated by X+, V+, and W+i+ sequentially from 1' to n (the number of characters on one page of the document of interest) from the bit image memory 103, and stores them in the character buffer 501 in the pattern matching circuit. At the same time, the size information w,, h, is transferred to the input buffer 50.
WA within 2.

Output to HA register. Then, the character generator 107 generates a character pattern for the first recognition candidate Cto for the character number i, and by scanning the 1゛° bit in the generated character pattern vertically and horizontally, the character pattern is Obtain the size information of the circumscribing rectangle. Then, the character pattern in the obtained circumscribed rectangle is transferred to one character buffer 503, and the size information of the circumscribed rectangle is set in the WB and HB registers in the input buffer 504. Although the explanation is complicated, when each character pattern is expanded into the character buffers 501 and 503, it is assumed that the upper left end of each buffer is used as a reference.

Now, when data is set in each character buffer 501, 503 and input buffer 502, 504 in this way,
The normalization circuit 507 expands and contracts the pattern of size WAXHA developed in the character buffer 501 into a pattern of size WNXHN and develops it in the buffer 508, and also expands and contracts the pattern of size WBXHB developed in the character buffer 503.
The pattern is expanded and contracted into the same <WNxHN pattern and developed in the buffer 509. Thereafter, the matrix matching circuit 510 matches the black pixels (the bit state is "
The sum SL of the number of overlapping white pixels (pixels whose bit state is ``0°'') and the number of overlapping white pixels (pixels whose bit state is ``0°'') and the ratio SL of the total number of pixels in the pattern area, 82. /
S2 is determined, and when it is smaller than a preset threshold T, a determination result flag 511 is set (set to 1°).

In summary, if a correct recognition result is obtained, the character patterns developed in the buffers 508 and 509 should be substantially the same, so in this case, Sl/S2 will be approximately "1". , on the other hand, in the case of erroneous recognition, the number of pixels common to each character pattern will be relatively small, resulting in a value close to "o".

Therefore, by checking whether the determination flag 511 is set, the CPU 11 can determine whether recognition has been performed correctly. Note that the value of this determination flag 511 is stored as is in the misrecognition flag 403 in the character information memory 106. In this way, the recognition process and matching process for all character patterns in the bit image memory 103 are completed.
When a table including the misrecognition flag 403 as shown in FIG. 4 is created, the process proceeds to step 521 in FIG.
Go to 0.

In step 5210, the data in the bit image memory 103 is transferred to the VRAM 116 and displayed on the display 114.
to be displayed. Then, all the characters for which the misrecognition flag 403 is set in the character information memory 106 are searched, and the size w is displayed at the display position (x+, y+) of each character.
h, is drawn by driving the graphic drawing circuit 117. This makes it possible to clearly indicate erroneously recognized characters to the operator.

Thereafter, the process advances to step 5207 to correct the misrecognized characters.

Various correction methods can be considered, but in the embodiment, the following methods were used.

First, the input unit 102 is operated to move the cursor to the character position drawn in the rectangular frame, and the coordinates are input (the obtained coordinates are assumed to be (xm, ym)).

Incidentally, the behavior of the cursor and the mechanism of the input unit 102 (mouse) are outside the scope of the present invention, so a description thereof will be omitted. Then the obtained coordinates (xm.

ym) by referring to the character position section 401 of the character information memory 106 to obtain the corresponding character number. At this time, if the coordinate position is outside the character rectangle, the corresponding character will not be found.
The coordinates are invalidated and input is awaited until the next coordinates are input or an instruction to end the character correction process is given in step 5207. Also, the input coordinates (xm, ym)
If the character number of the character indicated by is ``ni'', the candidate character code string CKo of the candidate character section 402 is
CK1. Rotation is performed such that the second candidate character Ckl of CK2... is at the beginning and C2° is at the end. A font pattern corresponding to the character code that has newly become the first candidate is generated by the character generator 107 and output to the pattern matching circuit 115 for matching processing. With this process, matching was successful (
If the judgment result flag 511 is off), the misrecognition flag m in the character information memory 106 is cleared, and the VRAM 11
Clear the rectangular frame of Xk+y*-wk+hk on 6. In this way, the operator repeats the same process until all rectangular frames are cleared. When all the characters have been corrected, send a correction end instruction to the input device 10.
2, and step 5207 is completed. Then, in step 8208, each of the current first candidate character codes CIO stored in the character information memory 106 is taken as the final recognition result, and the character codes from character number 1 to character number n are stored in the character code memory 109.

This completes the recognition of one page and moves on to the next page. However, the first recognized character code of the next page is the last recognized character code that was previously recognized and stored in the character code memory 110 It will be stored in order starting immediately after the character code.

Description of Second Embodiment (FIG. 6)> In the embodiment described above, the character pattern to be recognized and the character pattern to be pattern matched are fonts provided in the character generator 107 in advance. Therefore, if the font format of the recognition target and the font of the character generator are unlikely to match (for example, if the recognition target characters are Gothic and the character generator 107 is Mincho, or if the font design is slightly different) ), matrix smashing increases the amount of noise, which may make it difficult to determine erroneous recognition.

In the embodiment of Section 2, by acquiring two character patterns to be pattern matched from the recognition target, more accurate pattern matching with less noise is performed, and erroneous recognition is determined.

FIG. 6 is a block diagram of the character generator 107° in the second embodiment. Note that other components will be explained using FIG. 1.

In the illustration, the font memory 608 stores character patterns corresponding to character codes, similar to the font memory in a normal character generator.
In the embodiment, it is configured with a RAM, and in the initial state, no character pattern corresponding to any character code is stored yet. Information related to one character in the font memory 608 is stored in a flag column 609, a column 606 for storing the width of the stored character pattern, a column 610 for storing the height, and an area 60 for actually storing the character pattern.
It consists of 7 parts. The flag column 6o9 stores information as to whether the corresponding character pattern is valid or not, and in the initial state, as explained earlier, the character pattern is 1.
Since the flag is not always stored, the CPU 11 stores "o", which means invalid, in each flag column 609.

601 is a CG I10 buffer that inputs and outputs fonts, and 607 is a CG I10 buffer that inputs and outputs fonts.
This is a CG controller that controls the I10 buffer.

The operation will be explained below.

Similar to the processing in the first embodiment, the CPU I1 performs matching processing via the bus 113.
The character code of the character pattern to be extracted is stored in the character code buffer 602 in the 0 buffer 601. The CG controller 607 searches the font memory 608 for a character corresponding to that character code. For example, now the letter °°A′
” is stored in the character code buffer 602, the CG controller 607 checks whether the flag FA for the character “A°” is valid. If the flag FA is '1', that is, the character pattern is valid, the character width WA is set to CG I10 buffer 6.
The height HA is stored in the register (W) 603 in 01, and the height HA is stored in the same register (H) 604.
6o7 is transferred to pattern buffer 609. and,
CPU, 1jl is CG■〆0 buffer 6, . The register (W) 603 and register (H) in 01 are checked, and if the value is other than 0'', it is assumed that the corresponding pattern has been developed in the pattern buffer 609, and reading processing is performed.

On the other hand, when the valid flag FA is 0°, "o" is assigned to the register (W) 603 and the register (H) 604.

The above is the operation of extracting the font. Therefore, the processing contents in step 5209 in FIG. 2 will be explained as follows.

First, the character pattern to be recognized is transferred from the bit image memory 103 to the character buffer 501 in the pattern matching circuit 115 with reference to the character position section 401 of the character information memory 106 . Its size information is also input buffer 5
Transfer to 02. Next, the character generator 10
The character code is written to the character code buffer 602 of 7. By the above-described operation, the CG I10 buffer 60
1 register (W) 603. A value is set in register (H) 604. CPUI 11 is register (W) 60
3 (or register (H) 604) and enter “Ooo
If so, it is determined that a comparable font is not yet available, and the erroneous recognition flag 403 in the character information memory is set.
is set to °“1°°, and the pattern matching circuit 520
9 is not activated.

Further, when the register (W) 603 and register (H) 6044 have values other than "O°", the pattern buffer 604
Since the character pattern corresponding to the set character code is developed in 06, the character pattern is transferred to the character buffer 503 in the pattern matching circuit 115, and the contents of each register are input in the same way. The data is transferred to the buffer 504 and matching processing is started.

At this time, if the matching is successful, the misrecognition flag in the character information memory 106 is set to "o", and if it is unsuccessful, the misrecognition flag is set to 2°.

Next, in the process corresponding to step 5210, the erroneous recognition flag 403 in the character information memory is referred to, and a rectangular frame is drawn for the character whose erroneous recognition flag is determined to be "2°°" as in the previous embodiment. Display on the display.Also, for characters for which no font is available and pattern matching is not performed, that is, characters for which the misrecognition flag is "°1゛°," a dotted rectangular frame is drawn. , to inform you of this. Then, for characters that are determined to be correctly recognized and characters whose misrecognition flag is O°, nothing other than the character pattern is displayed. The operator corrects the characters in the rectangular frame by looking at the display, but the operator also checks whether the characters in the dotted frame are correct by comparing them to the context and the manuscript, and if the characters are incorrect. If so, make corrections (corresponding to step 5207).

Next, as in the first embodiment, a character code string is stored in the process corresponding to step 8208. At this time,
In the second embodiment, font addition processing is performed. Specifically, the final recognized character codes CIO for the characters numbered 1 to n in the character information memory 106 are transferred to the character generator 107'. As mentioned above, if the corresponding font is not available, register (
W) 603 and register (H) 604 contain °0°, and if the font has already been registered, O°' will be stored in both registers.

Therefore, the CPU III inputs character position part 4 only when register (W) 603 and register H 604 are “0°°”.
Corresponding character size information W r of 01.

hl is stored in the register (W) 603 and register (H) 604, and the character pattern is stored in the bit image memory 1.
03 and transfer it to the pattern buffer 606. After this, the CG controller 605
An instruction prompting for registration is input from , the character coat, width, height, and character pattern of the CG I10 buffer 601 are stored in a predetermined portion of the font memory 608, and the corresponding flag column 609 is set to 1°.

With the configuration described above, pattern matching can be performed using the same font as the one written on the original, so that extremely accurate recognition errors can be determined.

In particular, when the manuscript is in English, if the operator carefully processes only the first page, the character generator 10 can generate all character patterns of the alphabet.
This is extremely convenient since it will be constructed within 7 degrees.

In the second embodiment, the trigger for adding a font is page by page, but it may be added in smaller units such as line by line.

<Description of the third embodiment (Fig. 7)> Fig. 7 shows the character generator section 1 of the third embodiment.
07°', which includes a CG702 and a buffer T in the character generator 107' of the second embodiment described above.
The configuration includes YPE70. CO2O3 is a commonly used character generator (ROM
This is the same as that used in the first embodiment, and when a character code is sent, a preset character pattern is returned.

゛ In addition, in the second embodiment, the flag field in the font memory 608 is set to "If a character code of Ooo is sent, pattern matching is not performed and such a character is displayed on the display together with a dotted rectangle." However, in the third embodiment, if the font is not prepared in the font memory, the character code is sent to CO2O3 and the character pattern is
2O3, H604, and P2O3, and pattern matching is performed on the pattern. At this time, whether the character pattern developed in the pattern buffer 606 in the CG I10 buffer 601 was transferred from CO2O3,
Alternatively, information indicating whether the font has been transferred from the font memory 608 is set in the buffer TYPE 70. Specifically, when the character pattern in the font memory 608 is transferred to the pattern buffer 606, 1"' is transferred to the CO2
If it is from O3, set O″゛ as buffer TYP.
Assign '0'° to E701.

Then, as a result of pattern matching, judgment result flag 5
11 is set to 1°' and buffer TYPE7
If 01 is °1°', the misrecognition flag 403 is set to "2".
If the buffer TYPE 701 is "0°", the misrecognition flag 403 is set to "1°".
When the recognition result flag 511 is °'O', the misrecognition flag is set to °00°. Hereinafter, characters are displayed in the same manner as in the second embodiment.

As explained above, according to the third embodiment, as a result of pattern matching between character patterns in a document image,
Characters that are judged to have been misrecognized will be displayed in a rectangular frame, and characters judged to have been misrecognized as a result of pattern matching with the font that the device already has will be displayed in a dotted rectangular frame, so the operator can This can reduce the number of characters used to determine whether a file needs to be corrected or not.

In the embodiment, a frame is displayed for the incorrectly recognized character to distinguish it from other correctly recognized characters, but the present invention is not limited to this. This is because the same effect can be obtained by changing the display color, for example.

[Effects of the Invention] As explained above, according to the present invention, it is possible to determine whether a character pattern in a document image is a misrecognized character by matching it with a character pattern corresponding to a character code obtained by recognition. This has the effect of simplifying and reliably searching for and correcting misrecognized characters.

[Brief explanation of the drawing]

Fig. 1 is a block configuration diagram of the character recognition support device in this embodiment, Fig. 2 is a flowchart showing the processing procedure in this embodiment, Fig. 3 is a diagram showing character pattern cutout and its circumscribing rectangle, Fig. 4 is a diagram showing the contents of the character information memory in this embodiment, FIG. 5 is a diagram showing the configuration of the pattern matching circuit in this embodiment, FIG. 6 is a diagram showing the configuration of the character generator in the second embodiment, FIG. 7 is a diagram showing the configuration of a character generator in the third embodiment. In the figure, 101...image reading section, 102...input section, 103...bit image memory, 104...
Character extraction circuit, 105...Character recognition circuit, 106
...Character information memory, 107.107° and 107°
°...Character generator, 110...Character code memory, 111-CPU, 111 a-ROM
, 111b...RAM, 112...I10 interface, 113...system bus, 114...display, 115...pattern matching circuit, 11
6...VRAM, 117...Graphic drawing circuit, 501 and 503...Character buffer, 502 and 5
04... Input buffer, 507... Normalization circuit, 5
0.8 and 509... Buffer, 510... Matrix matching circuit, 511... Judgment result flag 51
1.601...CG I10 buffer, 605...
CG controller 608: font memory. ■ “

Claims (1)

[Scope of Claims] Recognition means for recognizing a character pattern to be recognized; pattern generation means for generating a character pattern corresponding to the character code obtained by recognition; and the generated character pattern and the character pattern to be recognized. a matching means for matching a character code, a determining means for determining whether or not the character code obtained by the recording recognition means is due to misrecognition based on the result of the matching means; 1. A character recognition support device comprising: notification means for notifying an external party that a corresponding character pattern to be recognized has been erroneously recognized.