JP2848560B2 - Image recognition method and apparatus - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image recognition method and apparatus for extracting an image block including a character pattern and recognizing the character.

2. Description of the Related Art Conventionally, in this type of apparatus, a character line is cut out by first taking a black pixel histogram in the vertical direction to cut out a character by optical character recognition. A circumscribed rectangle (hereinafter, referred to as a block) of a character is obtained by taking a horizontal histogram of the line. At this time, for example, "i", "ri", "le"
In the case of separated characters such as "Buddha", the block of each separated character is divided into two blocks on the left and right. When the combined block obtained by combining the left and right blocks has a width close to the average character width, one combined block is used. I try to cut it out.

Here, the character recognition operation will be described based on the schematic block configuration of the conventional character recognition device shown in FIG.

First, image data on a document optically read by the reading unit 51 is stored in the memory 52. Next, based on the image data, the block extraction unit 53 performs the above-described line extraction and block extraction. If there are blocks that can be combined, they are combined by the block combining unit 54 and stored in the character buffer 55. The recognition unit 56 recognizes the character standard pattern stored in the recognition dictionary unit 57 and the single block or the composite block stored in the character buffer 55. The standard pattern thus recognized is used as a recognition character, and its character code is stored in a word buffer (not shown) of the single collation unit 58. When a predetermined recognition character is stored in the word buffer, word matching with the word dictionary stored in the word dictionary unit 59 is performed.

[Problems to be solved by the invention] However, in the above conventional example, a period, a comma,
Since characters with a small character width, such as black and half-width numerals, are mixed, "3." and "." Are combined to form "3." as shown in FIG. You may make a mistake. For this reason, at the character extraction stage, correct character candidates are already eliminated. This can be compensated for by finding word candidates by comparing while allowing a difference of about one character at the word matching stage, but by losing the accuracy even for a single character, it is also possible to block Since the error in the above results in an error in the number of characters constituting a word, it is necessary to compare the similarity between words having different lengths, and there is a disadvantage that word matching becomes complicated.

The present invention has been made in view of the above conventional example,
When performing character recognition of an image block containing a character pattern,
It is determined whether or not the first image block of interest and the second image block adjacent thereto are to be combined. If it is determined that the first image block needs to be combined, the width of the third image block obtained by the combination is further determined. Then, the accuracy of the character width of the third image block is determined from the difference between the character width and the predetermined character width. Then, according to the accuracy, it is necessary to perform the synthesis by switching between character recognition of only the third image block and character recognition of the first image block and the second image block in addition to the third image block. There are image blocks,
It is another object of the present invention to provide an image recognition method and apparatus capable of more appropriately selecting the type of an image block that needs to recognize characters.

[Means for Solving the Problems] In order to achieve the above object, the image recognition method of the present invention includes the following steps. That is, it is necessary to input image information, extract an image block including a character pattern from the image information, and synthesize a first image block of interest in the extracted image block and a second image block adjacent thereto. It is determined whether or not it is necessary to combine the first image block and the second image block, the width of the third image block obtained by combining the first image block and the second image block, and a predetermined predetermined width The accuracy of the width of the third image block is obtained from the difference between the first image block and the third image block. And whether or not the second image block is also subjected to character recognition.

Further, in order to achieve the above object, the image recognition device of the present invention includes the following means. That is, input means for inputting image information, image block extracting means for extracting an image block including a character pattern from the image information input by the input means, and attention from among the image blocks extracted by the image block extracting means. Determining means for determining whether or not it is necessary to combine a first image block and a second image block adjacent thereto; and when the determining means determines that the first image block needs to be combined, the first image block An accuracy deriving means for obtaining the accuracy of the character width of the third image block from a difference between the width of the third image block obtained by combining the first image block and the second image block, and a predetermined character width; According to the accuracy obtained by the accuracy deriving means, only the third image block is character-recognized or the first image block and the third image block are added in addition to the third image block. Also the serial second image block to control so as to switch whether to character recognition and a character recognition control unit.

[Operation] With the above configuration, it is necessary to extract an image block including a character pattern from the image information, and combine the first image block of interest in the extracted image blocks with the adjacent second image block. It is determined whether or not there is. If it is determined that the third image block needs to be synthesized, the third image block obtained by synthesizing the first image block and the second image block is determined from the difference between the predetermined character width and the third image block.
Find the accuracy of the character width of an image block. Then, according to the accuracy, an operation is performed to switch between character recognition of only the third image block and character recognition of the first image block and the second image block in addition to the third image block.

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

<Description of First Embodiment> First, a first embodiment will be described.

FIG. 1 is a block diagram showing a first embodiment of the character recognition device according to the present invention. In the figure, reference numeral 1 denotes a character recognition device according to the first embodiment. Reference numeral 2 denotes a CPU for controlling the entire apparatus. Reference numeral 3 denotes a ROM storing a control program for operating the CPU 2, an error processing program, and programs according to flowcharts shown in FIGS. 3 (a) and 3 (b) which will be described later. A RAM used as a work area for various stored programs and a temporary save area during error processing is shown.
Reference numeral 5 denotes a reading unit that optically reads a document image, and reference numeral 6 denotes a memory that stores image data read by the reading unit 5. Reference numeral 7 denotes a block extracting unit for extracting a block of one character after extracting a character line from the image data stored in the memory 6, and 8 denotes a block and a block adjacent to each other in the blocks extracted by the block extracting unit 7. A block combining unit for performing block combining when the width (in this case, two block widths) is substantially equal to the average character width indicating the character width of one ordinary character.

Numeral 9 denotes a block accuracy judging unit for judging the accuracy of the synthesized block synthesized by the block synthesizing unit 8;
Indicates a character buffer for storing a block output based on the result of the determination by the block accuracy determination section 9 or a single block output without performing the determination processing in the block accuracy determination section 9. The number of blocks output to the character buffer 10 by the block accuracy judging section 9 is the number when the character is cut out once. Reference numeral 11 denotes a block counter for counting the total number of blocks in one character cut-out outputted by the block accuracy judging section 9. Reference numeral 12 denotes a recognition unit that performs character recognition by matching a standard pattern stored in a recognition dictionary unit 13 described later, which corresponds to the character pattern of the block stored in the character buffer 10. Reference numeral 13 denotes a recognition dictionary unit that stores standard patterns for matching.

Reference numeral 14 denotes a word matching unit that has a word buffer, enumerates word candidates based on the recognition result of the recognition unit 12, and performs word matching with a word dictionary stored in a word dictionary unit 15 described later. Reference numeral 15 denotes a word dictionary unit that stores a word dictionary to be matched with a word candidate as a recognition result.

Next, a character recognition method according to the first embodiment will be described.

2A and 2B are flow charts for explaining the operation of the CPU 2 of the first embodiment, and FIG. 3 is a diagram for explaining word matching of the first embodiment.

First, the original is optically read by the reading unit 5,
It is converted to a binary image and stored in the memory 6 (step
S1). Then, the block extracting section 7 takes a black pixel histogram in the main scanning direction, that is, the row direction, from the image data stored in the memory 6 and cuts out a character line using the valley position of the histogram as a character line cutout position ( Step
S2). With respect to the region of the character line thus cut out, the line is taken as a histogram of black pixels in the sub-scanning direction, that is, in a direction perpendicular to the line direction, and a circumscribed rectangle (hereinafter, referred to as a block) of each character soul is obtained. Ask (Step S3).

Next, the following processing procedure is sequentially performed between blocks. First, the total block width of the block width of the first first block and the block width of the next second block is calculated (step S4) and compared with a predetermined average character width set in advance. (Step S5). As a result, when it is determined that they are almost equal (for example, if the difference is within 20% of the error), it is determined that the two blocks should be combined, and the process proceeds to step 7. On the other hand, if the error exceeds 20%, it is determined that the two blocks should not be combined, and the first block is output to the block accuracy determining section 9 as a single block. In the case of this single block output,
The block accuracy judging unit 9 stores the data in the character buffer 10 without performing any processing. At the same time, the block counter 11
Is counted up by one, and the process proceeds to step S12 (step S6). Note that the initial value of the block counter 11 is "0".

Also, if the process proceeds from step S5 to step S7,
Two adjacent blocks are synthesized to generate one synthesized block (step S6). Then, the accuracy of block synthesis is obtained based on the following equation.

If the block accuracy calculated based on the above equation is smaller than 10%, that is, if the difference between the combined block width and the average character width is smaller than the average character width of 10%, it is determined that the block combination is accurate, On the other hand, if the block accuracy is 10% or more, it is determined that block synthesis is incorrect. Therefore, if it is determined that the composition of the blocks is incorrect, a single block is output to the character buffer 10 in the state of the blocks before the two blocks are composed (step S10), and two more blocks are composed. The synthesized block thus output is also output to the character buffer and the process proceeds to step S12 (step S11). In the above step S10, the block counter 11
Is counted up, and in the following step S11, the block counter 11 is counted up by one. Accordingly, the value of the block counter 11 becomes "3". If it is determined that the block synthesis is accurate, the process directly proceeds to step S11.
Proceed to step S11 to output only the synthesis block.
Proceed to. In this case, the value of the block counter 11 is "1".
Becomes

Here, in the example of FIG. 7 described in the related art,
For “ha” and “de”, the difference between the average character width is less than 10%, so no single block is output. For “3.”, the difference from the average character width is 10% or more, and the single block is output. Will be. Also, in the case of a block of characters other than ",", "," and "3.", no combination is performed and the block is output to the character buffer 10 as a single block. As described above, each character is output to the character buffer 10 including only a single block (step S6) and (2) to the character buffer 10 including only a composite block according to the presence / absence of block synthesis and the degree of determination of block synthesis. (Step S11), and (3) Output of a single block and a composite block to the character buffer (Step S10, Step S11), and character extraction is performed. In the above cases (1) and (2), one block is output to the character buffer 10. In the case of the above (3), three blocks (two single blocks and one composite block) are output to the character buffer 10.

Next, in step S12, the number of blocks corresponding to the value of the block counter 11 is extracted one by one from the character buffer 10, and character recognition is performed by a known technique using the recognition dictionary 13 in the recognition unit 12 (step S12). ). Then, the character code of the recognition result is sent to the word matching unit 14, and the word matching unit 15
Is stored in the word buffer. However, if the value of the block counter 11 is "3", the recognition result, that is, the control code is sent to the word buffer in the word collating unit 14 prior to sending the character code (step S14), and then 3 The two character codes are transmitted (step S15). Here, the recognition unit 12 according to each of the above-mentioned cases (1), (2), and (3).
Sends the recognition result. First, in the case of (1), one character code of the single block is transmitted, in the case of (2), one character code of the composite block is transmitted, and in the case of (3), the control code and the first code are transmitted. The character code of the single block, the character code of the second single block, and the character code of the composite block are transmitted. Where the character code is
JIS 2-byte codes are used, and unused 2-byte codes in the JIS code system are used as control codes.

In this way, the above-described processing is repeated until one word is stored in the word buffer in which word matching can be performed.
That is, once the transmission of the recognition result to the word buffer in step S15 is completed, it is determined in step S16 whether word collation is possible at that time. If it is determined in step S16 that it is impossible, it is first determined whether or not the next line needs to be cut out (step S17). While the process is not completed, the process returns to step S2 to repeat the above processing (step S1
8). If it is determined in step S17 that line segmenting is not necessary, the process returns to step S4 to repeat the above-described processing.

In this way, when one word by the character code (which may include a control code) is accumulated in the word buffer, the number of control codes in the word buffer is checked, and no control code is included. At this time, the content of the word buffer is output as it is as the final recognition result (step S1).
9, step S20). Therefore, the contents of the word buffer are output to the outside as a recognition result (step S27). If it is determined in step S19 that the control code is included in the word buffer, word matching is performed as follows. First, if there are three character codes following the control code,
1. The character code of the second block indicates the result of the first character recognition without block synthesis, and the character code of the composite block indicates the result of the second character recognition with block synthesis. Assuming that the word buffer contains n control codes under such conditions, the word candidates have 2 ⁿ combinations of character strings.

Here, the character string "alcohol" will be described as an example with reference to FIG. In this case, the above-mentioned (3) applies to the character of "R" as a result of character extraction. That is, ““ ”is represented by a control code and three character codes. Therefore, the character string "alcohol" has two characters "ru".
As a result, there are four types of word candidates.
That is, And four types of word candidates. Therefore, matching is performed for all the word candidates with the contents of the word dictionary unit 15, and the matching word candidates in the word dictionary unit 15 are set as final recognition results. Therefore, in the case of the above “alcohol”, word candidates (a), (b), and (c) indicate that the matching word is a word dictionary part.
Being rejected for not being within 15. In the last word candidate (d), since a matching word exists in the word dictionary section 15, this (d) Is the correct recognition result. Therefore, as the processing, word candidates are listed by the number (2 ⁿ ) of words to be collated in step S21,
Word matching is performed one by one while confirming the mismatch (steps S22 and S23). Here, if the next word candidate is cut off without a matching word, all the word candidates are regarded as the recognition result (step S24, step S2).
5), and outputs the recognition result to the outside (step S27).
When a matching word candidate is found in step S23, the matching word candidate is determined as a recognition result (step S2).
6). Then, output the recognition result to the outside (step S2
7).

Thus, after outputting the recognition result to the outside, the process returns to step S17 again, and the above-described processing is repeated.

As described above, according to the first embodiment, a plurality of character collations can be performed, and accurate character recognition can be performed by eliminating omissions in character recognition candidates.

<Description of Second Embodiment> Next, a second embodiment will be described.

FIG. 4 is a block diagram showing a second embodiment of the character recognition apparatus according to the present invention, and FIG. 5 is a block diagram showing the CPU 2 of the first embodiment.
3 is a flowchart for explaining the operation of FIG.

In FIG. 4, reference numeral 21 denotes a character recognition device according to the second embodiment. Reference numeral 22 denotes a CPU that controls the entire apparatus. Reference numeral 23 denotes a ROM which stores a control program for operating the CPU 22, an error processing program, and a program according to a flowchart shown in FIG. 5 which will be described later, and 24 denotes a program of various programs stored in the ROM 23. A RAM used as a work area and a temporary save area during error processing is shown. Here, since the function of each unit of the reference numerals 25 to 31 is the same as that of the above-described first embodiment, the description is omitted.

Then, 32 obtains a similarity by comparing the character pattern in the block stored in the character buffer 30 with a standard pattern stored in a recognition dictionary unit 33 described later, and based on the standard pattern having the maximum similarity. 6 shows a similarity calculation unit that lists recognition character candidates. Reference numeral 33 denotes a recognition dictionary unit that stores a standard pattern used by the similarity calculation unit 32 to list recognition character candidates. Reference numeral 34 denotes an identification unit that identifies the last recognition result from the recognition character candidates listed in the similarity calculation unit 32.

Here, a character recognition method according to the second embodiment will be described.

Also in the second embodiment, steps S1 to S1 of the first embodiment are performed.
The same processing as the processing up to step S11 is performed in steps S1 'to
This is performed up to step S11 '. Therefore, the description of the processing during that period is omitted.

Then, when a block is stored in the character buffer 30 by steps S1 'to S11', the value of the block counter 31 is checked next (step S30). If the value is "1", the similarity calculation unit 32 calculates the standard pattern and the similarity in the recognition dictionary unit 33 because the block is a single block or a composite block (step S31). Then, the standard pattern having the highest similarity is set as the recognition result (step S32). The recognition result obtained in this way is output to the outside without any processing in the identification section 34 (step S42). Also step S
When it is confirmed at step 30 that the value of the block counter 31 is "3", the similarity calculating section 32 checks each block (the first single block, the second single block, and the composite block) in the character buffer 30. Thus, the similarity with the standard pattern in the recognition dictionary unit 33 is calculated. Then determine the maximum similarity when each of the calculation, the largest similarity first single block, the maximum degree of similarity second single block, m ₁ a maximum similarity of the synthetic block, respectively, m _2, m ₃ (Step
S33 to step S38).

Next, the identification unit 34 identifies a standard pattern to be finally output as a recognized character from the maximum similarities m ₁ , m ₂ , and m ₃ .
The following equation is used as this identification method. That is, (m ₁ +
m ₂ ) If ÷ 2 ≧ m ₃ , the character code of the standard pattern having the highest similarity to the single block is determined as the final recognition result. In this case, the standard patterns respectively corresponding to the similarities m ₁ and m ₂ are identified as the recognition results (step S39, step S39).
S40). If (m ¹ + m ² ) ÷ 2 <m ³ , the final recognition result is the character code of the standard pattern having the highest similarity to the composite block. Standard pattern corresponding to the similarity m ₃ are identified as the recognition result in this case (step S41).

Next, step S32, step S40, and step S41
The recognition results identified by the respective identification processes are output to the outside (step S42). After outputting the recognition result thus identified to the outside, if it is not necessary to cut out the next line in order to perform the next character recognition, the process returns to the corresponding step S4 as in the first embodiment (step S43). If it is necessary to cut out the next line, the process returns to the corresponding step S2 as in the first embodiment until the processing for one page is completed, and the processing is repeated (step S44).

As described above, according to the second embodiment, even when character recognition is performed by similarity determination, accurate character recognition can be performed in a state where there is no omission in candidates for character recognition.

[Effects of the Invention] As described above, according to the present invention, when performing character recognition of an image block including a character pattern, it is determined whether or not to combine the first image block of interest and the second image block adjacent thereto. And if it is determined that it is necessary to combine,
Further, from the difference between the width of the third image block obtained by the synthesis and a predetermined character width, the third image block is obtained.
Find the accuracy of the character width of an image block. Then, according to the accuracy, it is necessary to perform the synthesis by switching between character recognition of only the third image block and character recognition of the first image block and the second image block in addition to the third image block. It is possible to provide an image recognition method and apparatus that can more appropriately select a certain image block and a type of an image block that requires character recognition.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a first embodiment of a character recognition apparatus according to the present invention. FIGS. 2 (a) and 2 (b) are flow charts for explaining the operation of the CPU 2 of the first embodiment. FIG. 4 is a diagram for explaining word matching according to the first embodiment. FIG. 4 is a block diagram showing a second embodiment of the character recognition device according to the present invention. FIG. 5 is an operation of the CPU 2 of the first embodiment. FIG. 6 is a block diagram showing a conventional character recognition device, and FIG. 7 is a diagram for explaining conventional word matching. In the figure, 1,21 ... character recognition device, 2,22 ... CPU, 3,23 ... R
OM, 4,24… RAM, 5,25,51… Reader, 6,26,52… Memory, 7,27,53… Block extractor, 8,28,54… Block combiner, 9,29 …… Block accuracy judgment section, 10,30,55…
… Character buffer, 11,31 …… Block counter, 12,56…
... Recognition unit, 13,33,57 ... Recognition dictionary unit, 14,58 ... Word collation unit, 15,59 ... Word dictionary unit, 32 ... Similarity calculation unit, 34
... An identification unit.

Claims (2)

(57) [Claims] An image information is input, an image block including a character pattern is extracted from the image information, and a first image block of interest in the extracted image block and a second image block adjacent thereto are synthesized. It is determined whether or not it is necessary. If it is determined that the image needs to be synthesized, the width of the third image block obtained by synthesizing the first image block and the second image block is set in advance. The accuracy of the width of the third image block is obtained from the difference from the predetermined width, and according to the accuracy, only the third image block is character-recognized or the third image block is added in addition to the third image block. An image recognition method, characterized in that whether or not one image block and the second image block are also subjected to character recognition is switched. 2. An input unit for inputting image information, an image block extracting unit for extracting an image block including a character pattern from the image information input by the input unit, and an image block extracted by the image block extracting unit. A first image block of interest, and a judging means for judging whether or not it is necessary to synthesize the second image block adjacent thereto. If the judging means judges that the first image block needs to be synthesized, Accuracy derivation for obtaining the accuracy of the character width of the third image block from the difference between the width of the third image block obtained by combining the one image block and the second image block and a predetermined character width. Means for character recognition of only the third image block, or the first image block in addition to the third image block, according to the accuracy obtained by the accuracy derivation means. And a character recognition control unit that controls whether or not the second image block also performs character recognition.