JP3093397B2 - Character recognition method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to character recognition, and more particularly to character recognition of multi-size, unfixed pitch Japanese documents.

[0002]

2. Description of the Related Art Many Japanese printed documents have variable character sizes or irregular pitches, and there are many characters such as kanji, hiragana, katakana, etc. which can be separated vertically and horizontally. It is difficult to cut out characters.

For this reason, a character block is cut out from the line image by taking a vertical projection or the like, and the character blocks are integrated to generate a character pattern that can be regarded as one character, and all generated character patterns are recognized. A method has been devised in which a character pattern is selected and cut out as a character by evaluating the likelihood of the result (for example, the distance from the recognition dictionary) (JP-A-1-303586).

[0004]

According to the conventional method of feeding back information of the recognition result to character segmentation, the character segmentation accuracy and recognition rate can be expected to be improved in character recognition for a multi-size, indefinite pitch Japanese document.
There is much room for improvement in the separation character extraction accuracy. For example, for a character that is separated side by side, a character pattern corresponding to the side or side and an original character pattern integrated with them are generated. A character pattern corresponding to a simple biased or adjacent character has a higher degree of certainty of the recognition result and is more likely to be selected as a character.

SUMMARY OF THE INVENTION It is therefore a primary object of the present invention to provide a character recognition method which is less likely to cause such an error in separation of separated characters and which is suitable for a multi-size, indefinite-pitch Japanese document.

[0006]

According to the gist of the present invention, a character pattern is cut out from a document image and collated with a character pattern dictionary for recognition. In advance, information on the number of blocks constituting a character is registered in the character pattern dictionary. In advance, the number of blocks constituting the cut-out character pattern is detected, and the detected number of blocks is compared with the number of blocks registered in the character pattern dictionary for the character as a recognition result. Then, for a character pattern in which the number of blocks does not match, the cutout method is changed and character cutout is performed again, and then the character pattern dictionary is collated and recognized.

[0007]

For example, when attention is paid to the vertical projection, the character pattern of "" can be divided into two blocks corresponding to the partial and adjacent sides. In other words, the "machine" is composed of two blocks. A block is a portion where such a character can be divided, and there are very many other characters that can be similarly divided into blocks.

Here, the blocks are defined by paying attention to the vertical projection. This is because the vertical projection makes it easy to detect the number of blocks of the character pattern cut out from the document. This is because it is considered that there are many characters to be separated into black and white. For example, it is also possible to make a block of connected black pixels. A “character block” in each embodiment described later is a block divided by vertical projection.

According to the present invention, a character pattern is cut out from an input image and recognized by collating it with a character pattern dictionary. The number of blocks registered in the character pattern dictionary for the character as a result of recognition is cut out. Compare the number of blocks of the character pattern. If there is no error in the number of blocks registered in the character pattern dictionary, the fact that the numbers of blocks match indicates that the possibility that characters have been cut out and recognized normally is sufficiently high.

By comparing the character cutout and the number of blocks as described above, it is less likely that a separation character is cut out incorrectly and that a mistaken recognition is not caused.

On the other hand, if the number of blocks does not match, there is a possibility that a cutout position is inappropriate and a cutout character is cut out as one character, and two half-width characters are cut out as one character. Is high. In this case, according to the present invention, since the character pattern is divided and each divided character pattern is recognized by comparing it again with the character pattern dictionary, inappropriate character cutout as described above is corrected and correct. The recognition result can be obtained.

[0012]

Embodiments of the present invention will be described below with reference to the drawings.

Embodiment 1 FIG. 1 is a block diagram of a character recognition apparatus according to this embodiment. Hereinafter, the processing content will be described according to the processing flow.

(Step 1) An original (document or form) is read as a binary image by a binary image input device 100 such as a scanner and stored in an image memory 102.

(Step 2) The line cutout unit 104 performs horizontal projection on the read image, and cuts out a section having a horizontal projection value exceeding a predetermined threshold and a certain width or more as a row image.

(Step 3) The character block extraction unit 106 performs vertical projection on the cut-out line image, and extracts a section where the vertical projection value exceeds a predetermined threshold value as a character block (corresponding to a block). . Character chunks whose space between adjacent chunks is less than a certain width are integrated. However, the width after integration should not exceed a certain value.

(Step 4) Standard character size calculation unit 10
8, the maximum value of the height of the character block in one line is determined, and this is set as the standard character size of the line.
10 is stored.

(Step 5) The character segmentation unit 112 combines the character chunks from the head of the line (here, the left end of the line because a horizontal writing document is considered) so as to have the maximum width without exceeding the standard character size. Each character block is cut out as a character pattern. The number of character chunks included in one character pattern naturally changes depending on the width and pitch of the character chunks. When the width of a character block or the margin between adjacent character blocks is large, the character block may naturally be cut out alone as a character pattern. Then, the number of character blocks, that is, the number of blocks constituting the cut-out character pattern is stored in the block number memory 114.

In the method described in JP-A-1-303586, the same character block may be used in a plurality of character patterns. However, in the present invention, differently, one character block is always cut out as a whole or a part of one character pattern, and the character blocks do not overlap between a plurality of character patterns.

(Step 6) In the recognition section 116, one character pattern extracted by the character extraction section 112 is collated with the character pattern dictionary 118, and a short candidate character is obtained as a recognition result. Usually, a plurality of candidate characters are obtained and sorted in ascending order of distance.

In the character pattern dictionary 118, information on the number of blocks is registered in addition to the characteristic amount of the character pattern. Here, the feature amount extracted from the extracted character pattern is compared with the feature amount registered in the character pattern dictionary 118.

(Step 7) The number-of-blocks reference section 120 reads the number of blocks of the currently focused character pattern from the number-of-blocks memory 114, and obtains the first pattern obtained by the recognition section 116 for the focused character pattern. The number of blocks of the position candidate character is read from the character pattern dictionary 118, and a check is made to see if the numbers of both blocks match.

If the number of blocks matches, data of the currently obtained recognition result (candidate character code, distance, etc.)
Is output to the recognition result memory 126, and the processing of one character is completed. That is, when the number of blocks matches, the character pattern of interest is cut out as a character.

On the other hand, if the number of blocks does not match,
The following steps 8 and later are performed.

(Step 8) If the target character pattern consists of two or more character blocks, it is divided into character blocks (blocks). If the character pattern consists of a single character block, for example, Force division at the position of the projection valley.

(Step 9) In the re-recognition unit 124,
In step 8, the character pattern re-cut (divided) is compared with the pattern dictionary 118 for recognition, and the recognition result data is stored in the recognition result memory 126.

Embodiment 2 FIG. 2 is a block diagram of a character recognition device according to this embodiment. 1 denote the same parts. Hereinafter, the processing content will be described according to the processing flow.

(Steps 1 to 4) Step 1 of Embodiment 1
The processing is the same as that of the processing of Steps 4 to 4.

(Step 5) In the same manner as in step 5 of the first embodiment, a character pattern is cut out by the character cutout unit 112A, and the number of blocks of each character pattern is stored in the block number memory 114. Character cutout unit 112A
Further, information on the combination of character blocks for each character pattern is stored in the combination information memory 130.

(Step 6) In the recognition unit 116A,
One character pattern extracted by the character extraction unit 112A is collated with the character pattern dictionary 118, and the code, distance, and total number of contours of the candidate character are stored in the recognition information memory 132 as a recognition result. The distance and the number of contours are used for evaluating the certainty of the recognition result as described later.

(Step 7) As in step 7 of the first embodiment, the number of blocks is compared, and if they match, the character pattern of interest is cut out as a character, and the recognition result is output to the recognition result memory 126. Processing of this character ends. If the block numbers do not match, the process proceeds to the next step 8.

(Step 8) The character cutout is repeated as in step 8 of the first embodiment. However, this re-cutout processing is executed by the character cutout unit 112A. Also, the combination information of the character blocks of the character pattern cut out here is stored in the combination information memory 130. Note that the character block combination information on the character pattern before re-cutout (before division) is also stored.

(Step 9) In the recognition unit 116A,
The character pattern re-cut out in step 8 is collated with the character pattern dictionary 118, and the recognition result data is stored in the recognition information memory 132. Note that the recognition result obtained in step 6 (the recognition result of the character pattern before being divided by re-cutting out the character) is also stored in the recognition information memory 132.

(Step 10) In the evaluation value calculating section 134, the first candidate of the recognition result before division obtained in step 6 and the first candidate of the recognition result after division obtained in step 9 are An evaluation value is calculated using the distance and the total number of contours and stored in the evaluation value memory 136. However, evaluation value = distance / total number of contours

(Step 11) Character pattern selector 13
In step 8, the maximum evaluation value of the recognition result after division is compared with the evaluation value of the recognition result before division, and the recognition result with the smaller value is output to the recognition result memory 126. That is, if the evaluation value of the recognition result before division is smaller, the character pattern before division is cut out as a character, otherwise, the character pattern after division is cut out as a character by re-cutting.

The contents of steps 8 to 11 will be described with reference to the example of the character string shown in FIG. In FIG. 3, a to f are character blocks directly extracted by vertical projection of a line image (before integration).
By performing an integration process on this, character chunks A to E are extracted.

With respect to the "machine" of this example, the character is cut out based on the standard character size in step 5 (2).
It is assumed that this character pattern is cut out, and this is recognized as "machine" in step 6. That is, it was correctly cut out and recognized.

However, for example, the character pattern dictionary 118
It is assumed that the block numbers do not match in step 7 due to the fact that the number of blocks for “machine” was incorrectly registered as 1.

Then, it is assumed that the "machine" is divided into two character patterns (1) and (3) in step 8, and these are recognized as "tree" and "how", respectively. According to the first embodiment, this is erroneously output as the final recognition result. However, according to this embodiment, (1) or (3)
If the evaluation value of any of the character patterns is larger than the evaluation value of the character pattern of (2) (this is usually the case), the character pattern of (2) can be selected and the recognition result can be output.

Similarly, the character pattern (5) is selected. The character pattern of (7) is cut out without going through re-cutting.

As described above, by introducing the evaluation of the certainty of the recognition result of the character pattern before and after the division, the number of blocks registered in the character pattern dictionary 118 is inappropriate, or 11
Since the difference between the character image used to create the character 8 and the character image to be actually recognized is large, even if the number of blocks does not match, it is possible to more reliably avoid separation character cutout errors. Conversely, in the case of two touching characters, each character separated by re-cutting can be cut out correctly.

Embodiment 3 FIG. 4 is a block diagram of a character recognition device according to this embodiment. 2 denote the same parts. The half-width character determination unit 140 and the character width memory 142 are newly provided. Differences between the processing contents of the present embodiment and the second embodiment are as follows.

First, the standard character size calculation unit 108B calculates a standard character width from the width of a character block in one line and stores the standard character width in the character width memory 142 separately from the standard character size.
The standard character width is a standard character width that can be regarded as a full-width character. For example, the maximum width of a character block in a line is used.

When the character pattern is divided by re-cutting out the character, the half-width character determination unit 140
The width of each character pattern after division is stored in the standard character width memory 142.
Compare with the standard character width read from
Character patterns having a width of / 2 or less are determined to be half-width characters. When the character pattern selection unit 138B recognizes a character pattern determined as a half-width character as a specific character type (for example, kanji or hiragana), the evaluation value is weighted (the evaluation value used here is , Larger means less certainty), making it difficult to select the character pattern after division, and comparing the evaluation value with the character pattern before division to select the character pattern to be finally cut out , And outputs the recognition result as a final recognition result.

By adding such processing, it is possible to prevent a separation character from being erroneously cut out as a half-width kanji or a half-width hiragana that has a low appearance frequency in a normal document.

Embodiment 4 FIG. 5 is a block diagram of a character recognition device according to this embodiment. 2 denote the same parts. The pitch information memory 146 is newly provided. Differences between the processing contents of the present embodiment and the second embodiment are as follows.

The character cutout unit 112C cuts out a character pattern on the basis of the standard character size. Apart from this, the space before and after each character block in the line is obtained and stored in the pitch information memory 146.

The character pattern selection section 138C switches the selection method between the character pattern after division by the character extraction again and the character pattern before division by pitch information.
That is, if the distance between the focused character block, which is the character pattern after division, and the next character block is larger than a value obtained by subtracting a certain value from the interval between the focused character block and the previous character block, Is likely to be a character by itself, and only between the character pattern consisting of the target character block and the character pattern integrating the target character block and the next character block,
The evaluation values are compared, and one of them is selected and cut out.
The method of selecting a character pattern when the above conditions are not satisfied is the same as in the second embodiment.

In the example shown in FIG. 6, when character extraction is performed again on the character pattern (4), attention is paid to a character block (3) which is one of the character patterns after division.
The interval b between this and the next character block (4) and the previous character block (2)
When the distance a between the character string satisfying the condition of b> ak (where k is a constant value), the comparison of the evaluation values is performed only between the character pattern (3) of the target character block alone and the character pattern (4). And select one of the character patterns. The character block 4, that is, the character pattern (7) is excluded from the target.

On the other hand, if the above condition is not satisfied, the maximum value of the evaluation values of the character patterns (3) and (7) is compared with the evaluation value of the character pattern (4) to obtain the character pattern (3). ) And (7), or the character pattern (4).

The constant value k is given as a fixed value or a value statistically obtained from the character pitch in one line.

Embodiment 5 FIG. 7 is a block diagram of a character recognition device according to this embodiment. 2 denote the same parts. The character size calculator 150 and the character size memory 152 are newly provided.

(Steps 1 to 3) Step 1 of Embodiment 2
In the same processing as in steps (1) to (3), a binary image input device
0 reads a document (document or form) into the image memory 102, cuts out a line image by the line cutout unit 104, and extracts a character block from the line image by the character block extracting unit 106.

(Step 4) Character pattern selection section 138
At D, the direction of the row is read. The line direction is specified in advance or automatically determined.

In the case of a horizontally written original, step 4 of the second embodiment
The horizontal writing process having the same contents as described below is performed, but in the case of a vertical writing document, the vertical writing process of the next step 5 and subsequent steps is executed.

(Step 5) Standard character size calculation unit 10
In step 8D, the standard character size in one line is calculated and stored in the standard character size memory 110. As a calculation method, a method of approximating the width of a line or a method of calculating the width of a character block in a line by statistical processing may be used.

(Step 6) In the same manner as in step 5 of the first embodiment, a character pattern is cut out by the character cutout unit 112A, the number of blocks of each character pattern is stored in the block number memory 114, and the character Combination information memory 1 by combining information of chunk combinations
30.

The character size of the extracted character pattern is obtained. Here, the character size is a value obtained by adding the pitch to the next character pattern to the width of the character pattern. And the difference between the character size and the standard character size (size difference)
Is stored in the character size memory 152.

(Step 7) In the recognition unit 116A,
The cut-out character pattern is collated with the character pattern dictionary 118, and the code, the distance, and the total number of contours of the candidate character are stored in the recognition information memory 132 as the recognition result.

(Step 8) Similar to step 7 of the first embodiment, the number of blocks is compared, and if they match, the character pattern of interest is cut out as a character, and the recognition result is output to the recognition result memory 126. Processing of this character ends. If the block numbers do not match, the process proceeds to the next step 9.

(Step 9) The character cutout unit 112A performs the same character cutout as in step 8 of the first embodiment.
The combination information of the character blocks of the extracted character pattern is stored in the combination information memory 130.
Character chunk combination information on the character pattern before re-cutout (before division) is also stored.

In the character size calculation section 150,
The character size is determined, the difference between the character size and the standard character size is calculated, and stored in the character size memory 152.

(Step 10) In the recognition section 116A, the character pattern re-cut out in step 9 is compared with the character pattern dictionary 118, and the recognition result data is stored in the recognition information memory 132. The recognition result of the character pattern before being divided by the character extraction again is also stored.

(Step 11) In the evaluation value calculation section 134, the first candidate of the recognition result before division obtained in step 7 and the first candidate of the recognition result after division obtained in step 10
For each candidate, an evaluation value is calculated using each distance and the total number of contours, and stored in the evaluation value memory 136. However, evaluation value = distance / total number of contours

(Step 12) Character pattern selector 13
In 8D, the size difference between the character patterns before and after the division is read from the character size memory 152, and the character patterns are sorted in ascending order of the size difference. In the example shown in FIG. 8, when D is a character pattern before division and is divided into C and E character patterns, these character patterns are sorted in the order of C, D, and E.

Since the size difference is used for such a purpose, by including the pitch with the next character pattern in the character size, a character pattern having an extremely small width such as A shown in FIG. To reduce the impact of being included.

(Step 13) Character pattern selector 13
8D reads the evaluation values of each character pattern before and after the division from the evaluation value memory 136, and sorts the character patterns in ascending order of the evaluation values.

(Step 14) Character pattern selector 13
In 8D, if the first character pattern sorted by the size difference and the first character pattern sorted by the evaluation value are the same, the character pattern is selected (finally as a character). The data of the recognition result is output to the recognition result memory 126. If not, a character pattern is selected in the next step 15.

(Step 15) If the difference between the evaluation values of the first and second character patterns sorted by the evaluation value is equal to or greater than a certain value, the first character pattern is selected, and the recognition result is obtained. Output. If not, the next step 16
To select a character pattern.

(Step 16) If the difference between the size difference between the first and second character patterns sorted by the size difference is equal to or greater than a certain value, the first character pattern is selected, and the recognition result is obtained. Output.

(Step 17) If the character pattern cannot be selected in any of steps 14 to 16, the first character pattern sorted by the evaluation value is selected.
The recognition result is output.

In the example shown in FIG. 8, when the character pattern of C, D, and E is selected, the character pattern is likely to be sorted in the order of C, D, and E due to the size difference, but is sorted by the evaluation value. Then D will be the first character pattern. Therefore, the character pattern of D is correctly selected in any of Steps 18, 16, and 17.

Embodiment 6 FIG. 9 is a block diagram of a character recognition device according to this embodiment. 2 denote the same parts. Differences between the processing contents of the present embodiment and the second embodiment are as follows.

The same processing as the processing of steps 1 to 10 of the second embodiment is performed. If the number of character blocks (blocks) included in the character pattern before being divided by re-cutout is two or less, character pattern selection processing having the same contents as in step 11 and subsequent steps of the second embodiment is performed.

However, when the character pattern before division is composed of three or more character blocks (blocks), many character patterns to be compared are generated, which takes time to process, and the risk of erroneous selection increases. Problem. Therefore, in this case, in the character pattern selection unit 138E, the comparison target is limited to three character patterns including the leading character block, and a character pattern having the smallest evaluation value among them is selected as a character. Output recognition result.

In the example of “river” shown in FIG.
There are three character patterns, but since the number of blocks of the character pattern c before division is 3, the evaluation value is limited to three character patterns a, b, and c including the leading character block. Are compared, and as a result, the character pattern of c is selected.

As described above, the character recognition apparatus according to the present invention uses a character pattern dictionary in which information on the number of character blocks is registered. This block number information is not included in the character pattern dictionary of the conventional character recognition device.

FIG. 11 is a schematic block diagram of a system for interactively registering the number of blocks manually in a prepared character pattern dictionary that does not include the number of blocks information. This system is incorporated in a character recognition device, or is constructed as a system independent of the character recognition device.

When registering the block number information in the character pattern dictionary, as preparation processing, data 202 of the character pattern dictionary already created except for the registration of the number of character blocks is prepared.
(Character codes and feature amounts, etc.) into a memory, and image data 20 used to create a character pattern dictionary.
4 is read into the memory, and an area 206 for storing the block number information is secured on the memory. Then, using the input device (keyboard, mouse, etc.) 208 and the display 210, the number of blocks is registered by the following interactive processing.

When the display of the character to be processed is designated from the input device 208, this instruction is notified to the dictionary creating unit 212 via the interactive processing unit 212, and the dictionary creating unit 212 converts the designated character of the character pattern dictionary data 202. The contents of the record and the image data of the designated character are stored in the interactive processing unit 212.
Is sent to the display 210 by the interaction processing unit 212. The operator displays 2
The character image is observed on the screen 10 and the number of blocks constituting the character is input from the input device 208. Dictionary creation unit 21
4 receives the input number of blocks via the interactive processing unit 212 and stores the information on the number of blocks in the corresponding position of the block number storage area 208 of the memory.

By such an interactive process, block number information is registered in a character pattern dictionary created in advance,
A character pattern dictionary required for character recognition according to the present invention can be created.

FIG. 12 is a block diagram of a system for creating a character pattern dictionary necessary for character recognition according to the present invention from the beginning. The dictionary creation process is as follows.

(Step 1) In the binary image input device 300, an original on which characters to be registered in the character pattern dictionary are printed is read by a scanner, converted into a binary image, and stored in the image memory 302. The original used at this time is printed at a constant pitch with a sufficiently wide character interval.

(Step 2) The line cutout unit 304 cuts out a line image from the binary image of the document by using vertical projection.

(Step 3) The character cutout unit 308 cuts out characters from the line image based on the print pitch of the original, extracts character chunks using vertical projection, and sets the character chunks forming each cutout character. Are stored in the block number memory 310 as the number of blocks.

(Step 4) The dictionary creation unit 312 extracts the characteristic amount of the cut-out character, and, together with this characteristic amount, extracts the number of blocks read from the block number memory 310 and the character image data read from the image memory 302. Register in the character pattern dictionary 314. The registration of the character image data in the character pattern dictionary 314 is for the convenience of a dictionary correction operation or the like.

By repeating the above steps 1 to 4, a character pattern dictionary 314 having a required number of characters can be created.

FIG. 13 is a block diagram of a system for correcting the character pattern dictionary created in this way. The outline of the correction process is as follows.

First, the character pattern dictionary 400 created by the system shown in FIG. 12 is read.

Input device (keyboard, mouse, etc.) 404
When the display of the character to be corrected is specified, this instruction is notified to the dictionary correction unit 402 via the interactive processing unit 408, and the dictionary correction unit 402
The image data of the designated character 0 and the block number information are read out, and are read out via the interactive processing unit 408 on the display 406.
Output to The operator checks the character image and the number of blocks displayed on the display 406, and inputs a correct number of blocks from the input device 404 if correction is necessary. The dialogue processing unit 408 compares the input block number with the previous block number, and if they do not match, passes the input block number to the dictionary correction unit 402 and instructs the dictionary correction unit 402 to correct it. The designated dictionary correction unit 402 rewrites the block number information of the corresponding character in the character pattern dictionary 400 with the passed block number.

[0091]

As can be understood from the above description, according to the first to seventh aspects of the present invention, attention is paid to the number of blocks constituting a character, the number of blocks of a recognized character pattern, and the recognition result. Character is compared with the number of blocks registered in the character pattern dictionary. If the number of blocks does not match, the character pattern is divided and recognized again, so high recognition is possible even for multi-size, undefined pitch Japanese documents. In particular, according to the third to seventh aspects of the present invention, it is possible to greatly improve the extraction accuracy and recognition rate for separated characters and the like in which character extraction is difficult.

[Brief description of the drawings]

FIG. 1 is a block diagram of a character recognition device according to a first embodiment.

FIG. 2 is a block diagram of a character recognition device according to a second embodiment of the present invention.

FIG. 3 is an explanatory diagram of a processing example according to a second embodiment.

FIG. 4 is a block diagram of a character recognition device according to a third embodiment.

FIG. 5 is a block diagram of a character recognition device according to a fourth embodiment.

FIG. 6 is an explanatory diagram of a processing example according to a fourth embodiment.

FIG. 7 is a block diagram of a character recognition device according to a fifth embodiment.

FIG. 8 is an explanatory diagram of a processing example according to a fifth embodiment.

FIG. 9 is a block diagram of a character recognition device according to a sixth embodiment.

FIG. 10 is an explanatory diagram of a processing example according to a sixth embodiment.

FIG. 11 is a block diagram of a system for registering block number information in a character pattern dictionary.

FIG. 12 is a block diagram of a system that reads a dictionary creation document and automatically creates a character pattern dictionary including information on the number of blocks of characters.

FIG. 13 is a block diagram of a system for correcting block number information of a character pattern dictionary.

[Explanation of symbols]

 Reference Signs List 100 Binary image input device 102 Image memory 104 Line segmentation unit 106 Character block extraction unit 108, 108B, 108D Standard character size calculation unit 110 Standard character size memory 112, 112A, 112C Character extraction unit 114 Block number memory 116, 116A Recognition unit 118 Character pattern dictionary 120 Number of blocks reference section 122 Re-character cutout section 124 Re-recognition section 126 Recognition result memory 130 Combination information memory 132 Recognition information memory 134 Evaluation value calculation section 136 Evaluation value memory 138, 138B to 138E Character pattern selection section 140 Half-width Character determination unit 142 Character width memory 146 Pitch information memory 150 Character size calculation unit 152 Character size memory

──────────────────────────────────────────────────続 き Continued on the front page (58) Field surveyed (Int.Cl. ⁷ , DB name) G06K 9/34 G06K 9/62

Claims (7)

(57) [Claims] 1. A character pattern is cut out from an input image.
The extracted character pattern is recognized by comparing it with a character pattern dictionary in which information on the number of blocks constituting the character is registered, and the number of blocks registered in the character pattern dictionary for the character as a result of recognition and the character pattern are determined. A character recognition method comprising: comparing the number of blocks constituting a character pattern; and, if the result of the comparison is inconsistent, dividing the character pattern into two or more character patterns and comparing the divided character pattern again with a character pattern dictionary for recognition. . 2. The character recognition method according to claim 1, wherein
A character recognition method characterized in that when a character pattern is divided and recognized again, a recognition result before or after division is selected as a final recognition result. 3. The character recognition method according to claim 2, wherein
A character recognition method comprising: selecting a final recognition result by comparing the likelihood of the recognition result after division with the certainty of the recognition result before division when the character pattern is divided and recognized again. 4. The character recognition method according to claim 3, wherein
When comparing the likelihood of the recognition results after division and before division, if the width of the character pattern after division can be regarded as half-width characters, and if the recognition result after division is a specific character type, A character recognition method characterized by lowering the degree of certainty of a later recognition result. 5. The character recognition method according to claim 3, wherein
A character recognition method characterized in that when comparing the likelihood of the recognition result after division and before the division, the method of comparing the likelihood of the recognition result is changed according to the interval before and after the character pattern after division. 6. The character recognition method according to claim 2, wherein
When the character pattern is divided and recognized again, based on the certainty of the recognition result after division and before division, and the value obtained by adding the pitch to the next character pattern to the width of the character pattern,
A character recognition method characterized by selecting a final recognition result. 7. The character recognition method according to claim 2, wherein
A character recognition method characterized by changing a method of selecting a final recognition result according to the number of blocks constituting a character pattern before division when a character pattern is divided and recognized again.