JP2009193170A - Character recognition device and character recognition method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To improve character recognition accuracy when there is fluctuation of an aspect ratio of a character or a difference of an aspect ratio by a character type or when characters are closed up or gradually written in smaller size. <P>SOLUTION: A line rectangle including characters is extracted from an image of a character recognition target (S201). In the line rectangle, pixel blocks are cut and divided (S202). The pixel blocks having vertical arrangement or the isolated pixel blocks among the cut and divided pixel blocks are combined, and the pixel block that is a unit of a character cut pattern is determined (S203, S204). Based on size of the determined pixel block, estimation character size of the notice pixel block is obtained from a local average (S205). The pixel blocks are combined such that the pixel blocks has size approximate to the obtained estimation character size (S207), and a permissible range is further widened to create a candidate of the character cut pattern (S208). Character recognition is performed about each obtained candidate, and a recognition result having most excellent evaluation is selected and output (S209). <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  The present invention relates to a character recognition device and a character recognition method for reading and recognizing characters entered on, for example, a paper document.

  There is a character recognition process in which an image recorded on a paper document is read by a scanner or the like and characters are automatically recognized from the read image. In the character recognition process, first, a pixel block (or pixel block) included in a recognition target region is cut out. Then, the segmented pixel block is cut out as it is or by combining a plurality of pixel blocks as one individual character pattern, the feature is extracted, and the extracted feature is stored in the recognition dictionary in which standard pattern information of the recognition target character is stored. To match. As a result of collation, the most probable character is selected based on matching features and the like, and the character code of the character pattern to be recognized is determined. Here, the image to be recognized is segmented into pixel blocks, and the character segmenting process for configuring individual character patterns to be recognized from the segmented pixel clusters is performed in the following procedure.

  In the first procedure, a plurality of pixel blocks delimited by blanks are cut out from a rectangular region including an image to be recognized. Then, the maximum value among the height of the row rectangle and the width of the cut out pixel block is set as the reference value. In order to combine pixel blocks and generate individual character patterns, one or a plurality of pixel blocks whose width is within (reference value + α) are recognized as one character pattern (for example, see Patent Document 1). .

In the second procedure, a character cutting method (referred to as a character cutting pattern) is not determined prior to recognition, and a plurality of character cutting patterns are created. A plurality of character cut patterns are sets of character cut patterns obtained by switching whether or not to combine adjacent pixel blocks. Then, for each character cutting pattern, a character is evaluated based on the character shape evaluation value obtained from the character shape of the character pattern cut out according to the character cutting pattern and the character recognition evaluation value obtained by comparing the candidate character pattern with the recognition dictionary. An evaluation value is calculated. The character cutting pattern with the maximum character evaluation value is the character cutting pattern for character recognition (see, for example, Patent Document 2).
JP 62-10784 A JP-A-3-191491

  However, in the first procedure described above, for example, when recognizing a handwritten address, the aspect ratio of the character is different depending on the writer, or the aspect ratio is different depending on the character type such as Chinese characters and numbers. For this reason, there is a drawback in that the original character is separated and recognized in the municipality and the town name part, and that a plurality of characters are originally extracted as one character in the chome address part. there were.

  In addition, when the margin of an entry field such as an address is reduced, the user tends to gradually write characters in a smaller size or gradually write smaller. For this reason, there is a drawback in that a plurality of characters that should originally be separated are cut out and recognized as a character obtained by combining them in a portion that is packed or small.

  Further, in the above-described second procedure, if a group of character cut patterns that covers all the combinations is created for the segmented pixel block, the number of character patterns to be evaluated becomes enormous. In particular, in the second procedure, it is necessary to evaluate the character shape evaluation value and the character recognition evaluation value of each character pattern for each of the plurality of character cutting patterns, and obtain the character evaluation value from those values, and the calculation amount is large. There was a problem of becoming enormous.

  The present invention has been made in view of the above conventional example, and has as its object to solve the above-described problems and improve the character recognition accuracy.

  In order to solve the above-described problems, the present invention has the following configuration.

An extraction means for extracting a region including a character string from image data composed of pixels;
A segmenting means for segmenting the region into pixel blocks which are units of combination;
Estimating means for sequentially paying attention to the pixel blocks segmented by the segmenting means and calculating the average width of the predetermined pixel clusters belonging to the local area including the target pixel cluster as the estimated character size of the target pixel cluster in the region When,
When the pixel blocks adjacent to each other are combined, if the width of the combined pixel block is within the first allowable range from the estimated character size, the segmentation is performed so that the combined pixel block becomes one pixel block. A character cutting means for creating a character cutting pattern in which the separation by means is changed,
Recognizing means for performing character recognition on a pixel block cut according to the character cutting pattern.

Alternatively, it is a character recognition device, and extraction means for extracting a region including a character string from image data constituted by pixels;
A segmenting means for segmenting the region into pixel blocks which are units of combination;
The relation between the position and width of the pixel block segmented by the segmenting means is approximated by a quadratic function, and the width corresponding to the position of each pixel block obtained by the quadratic function is used as the estimated character size of each pixel block. An estimation means to calculate;
When the pixel blocks adjacent to each other are combined, if the width of the combined pixel block is within the first allowable range from the estimated character size, the segmentation is performed so that the combined pixel block becomes one pixel block. A character cutting means for creating a character cutting pattern in which the separation by means is changed,
Recognizing means for performing character recognition on a pixel block cut according to the character cutting pattern.

  According to the present invention, the character recognition accuracy is improved. For example, even if there is a change in the aspect ratio of the character or the difference in the aspect ratio depending on the character type, or even if the character is stuffed or written gradually, it is misrecognized as a character that was originally separated. Or the possibility of misrecognizing the character as a combination of a plurality of characters. Also, there is an effect of reducing the amount of calculation by combining pixel blocks before creating a character cutting pattern that is a candidate for recognition.

[First Embodiment]
<Configuration of character recognition device>
FIG. 1 is a block diagram showing the configuration of a character recognition apparatus for carrying out the present invention. A CPU 101 controls the entire apparatus according to a control program stored in the ROM 102, and a ROM 102 stores a control program of the apparatus, such as processing shown in a flowchart described later, executed by the CPU 101. Reference numeral 103 denotes a RAM that stores form images and the like, and 104 denotes an external storage device such as a magnetic disk. Reference numeral 105 denotes a display, 106 denotes a keyboard, and 107 denotes a pointing device such as a mouse. An image scanner 108 reads a form image or the like.

  The present invention can also be implemented by a general-purpose computer. In that case, a computer program provided on a computer-readable storage medium or the like may be stored in the external storage device 104 and executed by the CPU 101 in accordance with an operator instruction or the like. Reference numeral 109 denotes a network interface, which communicates with a device located in a remote place (not shown) to read and write programs and data. When a general-purpose computer is used, an image scanner provided as a peripheral device is used as the image scanner 108.

<Character recognition processing>
Next, character recognition processing executed by the present embodiment will be described with reference to the flowchart of FIG. Although this description is given for a horizontally written document, the same method can be used for vertical writing after rotating the direction by 90 degrees. The procedure of FIG. 2 is realized by a program executed by the character recognition apparatus of FIG.

  First, row extraction is performed in step S201 from image data composed of pixels read by the image scanner 108. The image data to be extracted may be image data stored in the external storage device 104 or image data received from another image input device via the network interface 109. The row extraction may be performed by taking a projection in the row direction and taking a row rectangle so that the portion where the projection exists is the height. For example, a density histogram is obtained in the row direction, and an area having a height between a blank and a blank is extracted as a row. In other words, this region processing is processing for extracting a region including a character string from image data composed of pixels.

  Next, a group of continuous pixels (pixel block) is extracted from the row rectangle extracted in S202. For example, when the extracted pixel (also referred to as a row rectangle) is appropriately scanned to find a pixel that forms a character, and the contour is traced back to the original pixel, the pixel block is used as one pixel block. . Contour tracking is recorded, for example, as a sequence of tracked pixel coordinates. Then, the position and size of a row rectangle and a rectangle parallel to each side that circumscribes the pixel block are calculated. The position of the rectangle is indicated by a predetermined point such as the upper left corner. The size is indicated by, for example, the length of each side in the vertical and horizontal directions. A pixel block for which a circumscribed rectangle is once determined is indicated by the circumscribed rectangle. That is, after the pixel block is extracted in step S202, the pixel block is not a collection of pixels but a circumscribed rectangle. The output of the pixel block extraction processing in this embodiment is the position and size of the circumscribed rectangle of the pixel block. This is a processing convenience, and of course, the pixel collection itself can be stored and processed. In step S202, for example, a row rectangle is copied to a memory to extract a pixel block, and the extracted pixel block is deleted from the row rectangle. Then, pixel blocks are repeatedly extracted for the contents of the remaining row rectangles, and all the pixel blocks in the row rectangles are extracted.

  In step S203, the pixel blocks existing above and below are combined for horizontal writing. Combining pixel blocks means, for example, creating a minimum circumscribed rectangle including all circumscribed rectangles of pixel blocks to be combined as a new circumscribed rectangle after combining. The circumscribed rectangle of the pixel block to be combined is unnecessary and is deleted. As a result, for example, for a character having a crown as a radical, pixel blocks are combined to form an individual character pattern in units of one character. When there is a bias and a defect under the crown, one circumscribed rectangle circumscribing the three pixel blocks is obtained as a combined pixel block. However, some small pixel blocks of a character may be isolated and remain without being combined. In addition, it can be paraphrased that it exists mutually up and down about the direction orthogonal to the direction of a character string.

  In step S204, the pixel blocks existing close to the isolated pixel block are combined. The combination condition is, for example, that pixel blocks are adjacent to each other and the distance between adjacent pixel blocks is within a predetermined distance. This distance is not absolute and is not more than a predetermined number (for example, ¼) of the length of the short side of the smaller circumscribed rectangle. An additional condition such as a ratio between any of the vertical and horizontal sides being greater than a certain value (for example, 4 or more) or less than a certain value (for example, ¼ or less) can be added to this. Of course, this is only an example. To exclude punctuation marks, first recognize punctuation marks ("," and ".") From the position and shape of the pixel block, save the position, and then delete the pixel block corresponding to the punctuation mark from the line rectangle. You may do it.

  As described above, in steps S202 to S204, the pixel blocks to be combined are combined and divided into pixel blocks which are units to be combined in the subsequent steps. That is, an area including a character string to be recognized is cut into pixel blocks that are units to be combined for character recognition. In steps S203 and S204, when attention is paid to a certain pixel block, whether or not to combine them is determined only according to the relationship between the pixel of interest and its neighboring pixel block. In the pixel block combination in steps S206 and S208, which will be described later, the presence / absence of combination is determined using a value obtained by statistical processing, that is, the estimated character size. In this regard, steps S203 and 204 are steps S206 and S208. Is different.

  For the line rectangle, a histogram in a direction orthogonal to the character arrangement direction may be obtained, and characters may be cut at blank positions. In this way, the pixel block to be combined vertically is handled as one pixel block, so that the process of step S203 can be omitted.

  For example, from the row rectangle including “K” in FIG. 3, four pixel blocks 301 to 304 are extracted in step S202, and pixel blocks 301, 302, and 303 that overlap in the vertical direction are combined in step S203. A pixel block 305 is obtained. Next, in step S204, attention is paid to the pixel block 305 combined with the isolated pixel block 304. Since both pixels are close to each other and one pixel block is sufficiently smaller than the other, the small pixel block is separated from a part of the large pixel block. To be combined. Finally, an individual character pattern 306 is obtained.

  The pixel block after the combination of the top and bottom in step S203 and the isolated pixel block combination in S204 is a unit of character cut pattern combination. First, using this pixel block, an estimated character size determined by the position in the row direction is obtained in step S205. The character size is the character width in horizontal writing and the character height in vertical writing. The estimated character size of the pixel block of interest is, for example, a local average that is the average of the widths of the pixel block two pixels before and two pixels after the pixel block of interest. However, when there is no two previous pixel block or two subsequent pixel block, it is determined separately. For example, the estimated character size for the second pixel block from the top is the average of the first and fourth characters from the top, and the first estimated character size from the top is the second and the same value from the top. The estimated character size for the second pixel block from the end is the average of the first and fourth characters from the end, and the same applies to the first character from the end. Thereby, the estimated character size is determined for all the extracted pixel blocks. More generally, the estimated character size is not necessarily the average of the two pixel blocks described above. A local area of a certain size within a row rectangle including the pixel block of interest is defined, and an average value calculated from the size of the pixel block within the local area is set as the estimated character size. In the above example, an area including two pixel blocks before and after the pixel block of interest is defined as a local region, and an average of the sizes of the first and last pixel blocks included therein is used as the estimated character size. That is, in the row rectangle, attention is sequentially paid to the pixel blocks cut out in the procedure up to step S204, and the average of the widths of the predetermined pixel blocks belonging to the local region including the focused pixel block is calculated. The result is estimated as the estimated character size of the pixel block of interest. The estimated character size is a value for estimating the size of the pixel block of interest by referring to the size of the pixel block in the vicinity thereof, not the pixel block of interest. For example, when the size of the pixel block of interest is much smaller than the size of the estimated pixel block, it can be estimated that there is a high possibility that the pixel block of interest is not an independent character.

  FIG. 4 shows an example of an input character string. FIG. 5 shows the size (thick line) of each pixel block and the estimated character size (local average: thin line) of each pixel block with the position (row direction) of the pixel block as the X axis. ) On the Y-axis. For example, “Miyako” is divided into two pixel blocks of “person” 401 and “Ozama” 402 in FIG. Referring to FIG. 5, the actual size (width) of both is extremely small compared to the width of the surrounding pixel block (coordinate positions 501 and 502). However, referring to the estimated character size, it is larger than the neighboring pixel block (coordinate positions 511, 512). This indicates that there is a high possibility that the pixel blocks “person” and “Ozama” are combined to form one character. And that is true.

  When the estimated character size is obtained in this way, in step S206, the pixel blocks in the direction of the character string, that is, in the horizontal direction in horizontal writing, are combined using the estimated character size determined in step S205. However, there is a condition for combining, and if the size (width) of the combined pixel block is within the first allowable range based on the estimated character size of one of the pixel blocks to be combined, combining is performed. The estimated character size as a reference may be that of a predetermined pixel block, or may be the average of the estimated character sizes of the combined pixel block. For example, if the allowable range of the character width is 0.9 to 1.1 times the estimated character size and the character width when the two pixel blocks are combined is within the allowable range, the two are combined. The connection here is the same as in step S203. Since the segmentation of the pixel block has already been determined, it can be said that the process of creating the character segmentation pattern by changing the segmentation of the existing pixel block in step S206.

  In step S206, if the condition is not satisfied, the pixel block is not combined, so the combination does not necessarily occur. If left-right combination occurs in step S206, the process returns to the estimated character size calculation step in step S205 using the pixel block information after combination in step S207. For example, the position and size of the circumscribed rectangle of the pixel block before combining are replaced with the position and size of the circumscribed rectangle of the combined pixel block, and the calculation of the estimated character size is repeated again. If left-right coupling does not occur, the process proceeds to step S208.

  In step S208, a lattice structure is generated by combining the pixel blocks obtained in the steps so far. That is, based on the character segmentation pattern indicating the segmentation of the pixel block obtained up to step S207, the segmentation of the pixel block is changed so as to combine the adjacent pixel clusters and a new character segmentation pattern is generated. This is not a substitute for the original character cut pattern, but is generated additionally. The created character cutting pattern is a candidate for a character cutting pattern to be finally determined. The pixel block combination is again performed under certain conditions. That is, if the size (width) of the combined pixel block is within the second allowable range based on the estimated character size of one of the pixel blocks to be combined, the combined pixel block is combined. The second allowable range is wider than the first allowable range. For example, a larger one that is 1.2 times the estimated character size or 1.2 times the character line height is set as the second allowable range. If the combined width of the pixel blocks is within an allowable range (or threshold), the combined character cutting pattern is added to the character cutting pattern candidate (ie, lattice structure). In step S208, one or more candidates for all character cutting patterns that meet the conditions are created.

  In step S209, character recognition is performed on the pixel block cut according to each of the character cutting patterns created in step S208. Then, for each character cutting pattern, the sum of evaluation values obtained as a result of character recognition for each character is calculated. A total value of the evaluation values is obtained for each of the character cutting pattern candidates, and compared with each other to select a character cutting pattern that maximizes the evaluation value (that is, most likely). A character string recognized according to the selected character cutting pattern is output as a recognition result. The character recognition process itself may be a known method. For example, the feature registered in the identification dictionary is compared with the feature extracted from the pixel block to be recognized, and the character with the most matching points is output as the recognition result. To do. The evaluation value is high if there are many matching points of features.

  Through the above procedure, character recognition is performed for each of a plurality of character cut pattern candidates, and the result is evaluated, and a character string recognized according to the character cut pattern determined to be most appropriate is used as an output result. Further, since the character cut patterns that are candidates are limited to character cut patterns that have been statistically determined to be highly probable, the processing amount can be considerably smaller than that of the conventional example.

  As described above, according to the present embodiment, there is an effect that an estimated character size can be obtained with a small error from the local average even when characters are separated into left and right. In addition, since the local average is used instead of the average of the whole character string, even when the user gradually writes characters in small letters or writes them in small letters, it is possible to cope with higher accuracy. Further, by repeating the above processing, even if characters separated on the left and right are continuous, there is an effect that an individual character pattern can be gradually created and finally all individual character patterns can be extracted.

[Modification]
In this embodiment, after step S207, the process proceeds to step S208. However, the process may proceed to step S209. In this case, since a plurality of character cutting pattern candidates are not created, a recognition result estimated to be appropriate is not selected from the candidates, but the processing amount is greatly reduced. In addition, in the process of step S207, since a character cutting pattern with a statistically high possibility is determined, it can be expected that the recognition result is relatively good.

[Second Embodiment]
Next, a character recognition process executed by the second embodiment different from the above-described embodiment, which is executed by the character recognition device shown in FIG. 1, will be described with reference to the flowchart shown in FIG. In the present embodiment, FIG. 2 is replaced by the procedure of FIG.

<Character recognition processing>
Next, a character recognition process executed by the present embodiment will be described with reference to the flowchart of FIG. Although this description is given for a horizontally written document, the same method can be used for vertical writing after rotating the direction by 90 degrees. The procedure of FIG. 6 is realized by a program executed by the character recognition device of FIG.

  First, line extraction is performed in step S601 from image data composed of pixels read by the image scanner. The image data to be extracted may be image data stored in the external storage device 104 or image data received from another image input device via the network interface 109. The row extraction may be performed by taking a projection in the row direction and taking a row rectangle so that the portion where the projection exists is the height. For example, a density histogram is obtained in the row direction, and an area having a height between a blank and a blank is extracted as a row. In other words, this region processing is processing for extracting a region including a character string from image data composed of pixels.

  Next, a pixel block (pixel block) is extracted from the row rectangle extracted in S602. The pixel block is, for example, a pixel that forms a character by appropriately scanning the extracted area (also referred to as a row rectangle) and tracing the outline back to the original pixel. To do. Contour tracking is recorded, for example, as a sequence of tracked pixel coordinates. Then, the position and size of a row rectangle and a rectangle parallel to each side that circumscribes the pixel block are calculated. The position of the rectangle is indicated by a predetermined point such as the upper left corner. The size is indicated by, for example, the length of each side in the vertical and horizontal directions. A pixel block for which a circumscribed rectangle is once determined is indicated by the circumscribed rectangle. That is, after the pixel block is extracted in step S602, the pixel block is not a collection of pixels but a circumscribed rectangle. The output of the pixel block extraction processing in this embodiment is the position and size of the circumscribed rectangle of the pixel block. This is a processing convenience, and of course, the pixel collection itself can be stored and processed. In step S602, for example, the pixel rectangle is extracted by copying the row rectangle into the memory, and the extracted pixel block is deleted from the row rectangle. Then, pixel blocks are repeatedly extracted for the contents of the remaining row rectangles, and all the pixel blocks in the row rectangles are extracted.

  In step S603, the pixel blocks existing above and below are combined for horizontal writing. Combining pixel blocks means, for example, creating a minimum circumscribed rectangle including all circumscribed rectangles of pixel blocks to be combined as a new circumscribed rectangle after combining. The circumscribed rectangle of the pixel block to be combined is unnecessary and is deleted. As a result, for example, for a character having a crown as a radical, pixel blocks are combined to form an individual character pattern in units of one character. When there is a bias and a defect under the crown, one circumscribed rectangle circumscribing the three pixel blocks is obtained as a combined pixel block. However, some small pixel blocks of a character may be isolated and remain without being combined.

  In step S604, the pixel blocks existing close to the isolated pixel block are combined. The combination condition is, for example, that pixel blocks are adjacent to each other and the distance between adjacent pixel blocks is within a predetermined distance. This distance is not absolute and is not more than a predetermined number (for example, ¼) of the length of the short side of the smaller circumscribed rectangle. An additional condition such as a ratio between any of the vertical and horizontal sides being greater than a certain value (for example, 4 or more) or less than a certain value (for example, ¼ or less) can be added to this. Of course, this is only an example. To exclude punctuation marks, first recognize punctuation marks ("," and ".") From the position and shape of the pixel block, save the position, and then delete the pixel block corresponding to the punctuation mark from the line rectangle. You may do it.

  As described above, in steps S602 to S604, the pixel blocks to be combined are combined and divided into pixel blocks which are units to be combined in the subsequent steps.

  The pixel block that has undergone the vertical combination in step S603 and the isolated pixel block combination in step S604 is a unit of character cut pattern combination. Using this pixel block, an estimated character size determined by the position in the row direction is obtained in step S605. The estimated character size in step S605 is determined by an approximate quadratic function that gives the corresponding width by parameterizing the position of the pixel block in the row direction. This approximate quadratic function is determined so that, for example, the sum of squares of the distances from each point indicating the actual position and width of each pixel block is minimized. That is, the approximate quadratic function can be uniquely determined by determining three coefficients by the least square method.

  The input character string is shown in FIG. 4 as in the first embodiment. FIG. 7 is a graph obtained by approximating the size of a pixel block with a quadratic function with the row direction taken as the X-axis. Also in FIG. 7, referring to the estimated character sizes of the two pixel blocks “person” 401 and “Ozama” 402, it can be seen that the pixel blocks are in the vicinity. This indicates that there is a high possibility that the pixel blocks “person” and “Ozama” are combined to form one character.

  When the estimated character size is obtained in this way, in step S606, pixel blocks in the direction of the character string, that is, in the horizontal direction in horizontal writing, are combined using the estimated character size determined in step S605. However, there is a condition for combining, and if the size (width) of the combined pixel block is within the first allowable range based on the estimated character size of one of the pixel blocks to be combined, combining is performed. The estimated character size as a reference may be that of a predetermined pixel block, or may be the average of the estimated character sizes of the combined pixel block. For example, if the allowable range of the character width is 0.9 to 1.1 times the estimated character size and the character width when the two pixel blocks are combined is within the allowable range, the two are combined. Since the segmentation of the pixel block has already been determined, it can be said that the process of creating the character segmentation pattern by changing the segmentation of the existing pixel block in step S606.

  In this embodiment, since the estimated character size is approximated by a quadratic function, it is not necessary to recalculate the estimated character size even if pixel blocks are combined. In step S607, a lattice structure is generated by combining the pixel blocks obtained in the steps so far. That is, based on the character segmentation pattern indicating segmentation of pixel blocks obtained up to step S606, the segmentation of pixel blocks is changed so as to combine adjacent pixel clusters, and a new character segmentation pattern is generated. This is not a substitute for the original character cut pattern, but is generated additionally. The created character cutting pattern is a candidate for a character cutting pattern to be finally determined. The pixel block combination is again performed under certain conditions. That is, if the size (width) of the combined pixel block is within the second allowable range based on the estimated character size of one of the pixel blocks to be combined, the combined pixel block is combined. For example, a larger one that is 1.2 times the estimated character size or 1.2 times the character line height is set as the second allowable range. If the combined width of the pixel blocks is within an allowable range (or threshold), the combined character cutting pattern is added to the character cutting pattern candidate (ie, lattice structure). In step S607, all character cutting pattern candidates that meet the conditions are created.

  In step S608, character recognition is performed on the pixel block cut according to each of the character cut patterns created in step S607. Then, for each character cutting pattern, the sum of evaluation values obtained as a result of character recognition for each character is calculated. A total value of the evaluation values is obtained for each character cutting pattern candidate and compared with each other, and a character cutting pattern that maximizes the evaluation value is selected. A character string recognized according to the selected character cutting pattern is output as a recognition result.

  Through the above procedure, character recognition is performed for each of a plurality of character cut pattern candidates, and the result is evaluated, and a character string recognized according to the character cut pattern determined to be most appropriate is used as an output result. Further, since the character cut patterns that are candidates are limited to character cut patterns that have been statistically determined to be highly probable, the processing amount can be considerably smaller than that of the conventional example.

  As described above, according to the present embodiment, there is an effect that an estimated character size can be obtained with a small error from the local average even when characters are separated into left and right. Further, by repeating the above processing, even if characters separated on the left and right are continuous, there is an effect that an individual character pattern can be gradually created and finally all individual character patterns can be extracted.

  According to the present embodiment, there is an effect that the tendency of the character size can be appropriately estimated and approximated even when the character is gradually reduced. Further, even if pixel blocks are combined, it is not necessary to recalculate the estimated character size, and the processing load is reduced.

[Modification]
In this embodiment, after step S606, the process proceeds to step S607. However, the process may proceed to step S608. This is the same as in the first embodiment.

[Other Embodiments]
Note that the present invention can be applied to a system (for example, a copier, a facsimile machine, etc.) consisting of a single device even if it is applied to a system composed of a plurality of devices (eg, a host computer, interface device, reader, printer, etc.). You may apply. Another object of the present invention is to supply a recording medium recording a program code for realizing the functions of the above-described embodiments to a system or apparatus, and the system or apparatus computer reads out and executes the program code stored in the storage medium. Is also achieved. In this case, the program code itself read from the storage medium realizes the functions of the above-described embodiments, and the program code itself and the storage medium storing the program code constitute the present invention.

  Further, according to the present invention, the operating system (OS) running on the computer performs part or all of the actual processing based on the instruction of the program code, and the functions of the above-described embodiments are realized by the processing. This is also included. Furthermore, the present invention is also applied to the case where the program code read from the storage medium is written in a memory provided in a function expansion card inserted into the computer or a function expansion unit connected to the computer. In that case, based on the instruction of the written program code, the CPU of the function expansion card or the function expansion unit performs part or all of the actual processing, and the functions of the above-described embodiments are realized by the processing. .

It is a figure which shows the structure of the character recognition apparatus which concerns on Embodiment 1-2 of this invention. It is a flowchart for demonstrating the character recognition process which concerns on Embodiment 1 of this invention. It is a figure explaining the combination process of the character block which concerns on Embodiment 1 or 2 of this invention. It is a figure which shows the input character string image which concerns on Embodiment 1-2 of this invention. It is a graph which shows the estimated character size which concerns on Embodiment 1 of this invention. It is a flowchart for demonstrating the character recognition process which concerns on Embodiment 2 of this invention. It is a graph which shows the estimated character size which concerns on Embodiment 2 of this invention.

Explanation of symbols

101 ... CPU
102 ... ROM
103 ... RAM
104 ... External storage device 105 ... Display 106 ... Keyboard 107 ... Pointing device 108 ... Image scanner 109 ... Network interface

Claims (9)

Extraction means for extracting a region including a character string from image data composed of pixels;
A segmenting means for segmenting the region into pixel blocks which are units of combination;
Estimating means for sequentially paying attention to the pixel blocks segmented by the segmenting means and calculating the average width of the predetermined pixel clusters belonging to the local area including the target pixel cluster as the estimated character size of the target pixel cluster in the region When,
When the pixel blocks adjacent to each other are combined, if the width of the combined pixel block is within the first allowable range from the estimated character size, the segmentation is performed so that the combined pixel block becomes one pixel block. A character cutting means for creating a character cutting pattern in which the separation by means is changed,
A character recognition apparatus comprising: a recognition unit configured to perform character recognition on a pixel block cut according to the character cut pattern. Extraction means for extracting a region including a character string from image data composed of pixels;
A segmenting means for segmenting the region into pixel blocks which are units of combination;
The relation between the position and width of the pixel block segmented by the segmenting means is approximated by a quadratic function, and the width corresponding to the position of each pixel block obtained by the quadratic function is used as the estimated character size of each pixel block. An estimation means to calculate;
When the pixel blocks adjacent to each other are combined, if the width of the combined pixel block is within the first allowable range from the estimated character size, the segmentation is performed so that the combined pixel block becomes one pixel block. A character cutting means for creating a character cutting pattern in which the separation by means is changed,
A character recognition apparatus comprising: a recognition unit configured to perform character recognition on a pixel block cut according to the character cut pattern. When the pixel blocks whose separation has been changed by the character cutting means that are adjacent to each other are combined, the width of the combined pixel block is larger than the first allowable range based on the estimated character size. If it is within the range, it further comprises candidate creating means for creating one or a plurality of character cutting pattern candidates whose cutting by the character cutting means is changed so that the combined pixel block becomes one pixel block,
The recognizing unit performs character recognition on a pixel block cut in accordance with each of the character cutting pattern created by the character cutting unit and the character cutting pattern candidate created by the candidate creating unit, and each character cutting pattern The character recognition apparatus according to claim 1, wherein the most likely recognition result among the recognition results corresponding to is output as a character recognition result.   The segmenting means segments a group of continuous pixels as a pixel block, pays attention to the segmented pixel block in sequence, and overlaps in a direction perpendicular to the direction of the character string and the pixel block within a predetermined distance from the target pixel block 4. The character recognition device according to claim 1, wherein a pixel block to be processed is cut out as a pixel block as a unit of the combination. 5. A character recognition method executed by a character recognition device comprising an extraction means, a separation means, an estimation means, a character cut means, and a recognition means,
An extracting step in which the extracting means extracts a region including a character string from image data composed of pixels;
The dividing means for dividing the area into pixel blocks that are units of combination; and
The estimation means sequentially pays attention to the pixel blocks cut out by the cutting step, and within the region, the average width of predetermined pixel blocks belonging to the local area including the target pixel block is calculated as the estimated character size of the target pixel block. An estimation process to calculate as
When the character cutout unit combines the pixel blocks adjacent to each other, if the width of the combined pixel block is within the first allowable range from the estimated character size, the combined pixel block is one pixel block. A character cutting step for creating a character cutting pattern in which the cutting by the cutting means is changed,
A character recognition method comprising: a recognition step in which the recognition means performs character recognition on a pixel block cut according to the character cut pattern. A character recognition method executed by a character recognition device comprising an extraction means, a separation means, an estimation means, a character cut means, and a recognition means,
An extracting step in which the extracting means extracts a region including a character string from image data composed of pixels;
The dividing means for dividing the area into pixel blocks that are units of combination; and
The estimation means approximates the relationship between the position and width of the pixel block segmented by the segmenting step by a quadratic function, and sets the width corresponding to the position of each pixel block obtained by the quadratic function to each pixel block. An estimation process to calculate the estimated character size of
When the character cutout unit combines the pixel blocks adjacent to each other, if the width of the combined pixel block is within the first allowable range from the estimated character size, the combined pixel block is one pixel block. A character cutting step for creating a character cutting pattern in which the cutting by the cutting means is changed,
A character recognition method comprising: a recognition step in which the recognition means performs character recognition on a pixel block cut according to the character cut pattern. The character recognition device further includes candidate creation means,
When the candidate creation means combines the pixel blocks whose separation has been changed by the character cutting step adjacent to each other, the width of the combined pixel block is determined from the estimated character size and the first allowable range. A candidate creation step for creating one or a plurality of character cut pattern candidates in which the segmentation by the character cutting step is changed so that the combined pixel block becomes one pixel block within a second allowable range wider than Further comprising
In the recognizing step, character recognition is performed on a pixel block cut in accordance with each of the character cutting pattern created in the character cutting step and the character cutting pattern candidates created in the candidate creating step, and each character cutting pattern 7. The character recognition method according to claim 5, wherein the most probable recognition result among the recognition results corresponding to is output as a character recognition result.   In the segmentation step, a group of consecutive pixels is segmented as a pixel block, and attention is sequentially paid to the segmented pixel blocks, overlapping each other in a direction perpendicular to the direction of the character string and the pixel block within a predetermined distance from the target pixel block The character recognition method according to any one of claims 5 to 7, wherein a pixel block to be processed is cut out as a pixel block as a unit of the combination.   A program for causing a computer to function as the character recognition device according to any one of claims 1 to 4.

Images