JP6523988B2 - Character recognition device, character recognition method, and program - Google Patents

Description

  Embodiments of the present invention relate to a character recognition device, a character recognition method, and a program.

  A character recognition technology that automatically recognizes optically read characters is widely used. Character recognition processing is roughly classified into handwritten character recognition and printed character recognition. Here, the printed character recognition includes not only characters printed by printed characters, but also recognition of printed characters based on fonts by electronic data. The character recognition processing is performed by collating the input character image with the character data contained in the dictionary and evaluating a score (or distance) indicating the degree of matching. The character data contained in the dictionary is information representing the character form itself or the feature in recognition. Therefore, for example, for a certain character (for example, the character “a”), the data of the character included in the dictionary differs for each font. In the print character recognition process, character recognition is performed using dictionary data prepared in advance for each font.

In the prior art, there has been a technique of calculating an average value of minimum distance values between dictionary data of several characters and input data, and selecting a font having the minimum average value. However, the selected few characters may not necessarily match the font switching unit.
Further, in the prior art, there has been a technique of counting the number of times that the distance between the dictionary and the input character exceeds a predetermined threshold, and switching the dictionary based on the count value. However, there have been cases where the location where the dictionary is switched based on the count value does not match the location of font switching.
Further, in the prior art, there is a possibility that the process of determining the font becomes redundant, such as determining the font used in units of one character.
Further, in the prior art, there has been a technique of judging the most suitable font data (dictionary) and performing character recognition using the font data selected as a result. However, if the font changes in the middle of a sentence, it may not be optimal in the entire selected sentence.

Japanese Patent Application Laid-Open No. 5-159107 JP, 2006-092138, A Unexamined-Japanese-Patent No. 1-180083 Japanese Patent Application Laid-Open No. 63-238688 Unexamined-Japanese-Patent No. 3-068092

  The problem to be solved by the present invention is a character recognition apparatus and character recognition method capable of accurately detecting font change points in a document and using a dictionary suitable for the font being used. , And to provide the program.

  The character recognition device according to the embodiment has a recognition database, a character recognition unit, a font selection unit, and a selection reconsideration determination unit. The recognition database holds dictionary data for character recognition for each font type. The character recognition unit acquires a string of character images, performs recognition processing of the character image while referring to the dictionary data in the recognition database, and performs the recognition process on the dictionary data and the character image. Output a score representing the degree of match between The font selection unit is configured to specify a font type of the character image based on the character image and the dictionary data for each font type in the recognition database. Set the type. The selection reconsideration determination unit detects a section in the row in which the score is lowered when the character recognition unit sequentially performs the recognition process of the character image included in the string of the character image, and also in the rule base. The font change point in the section or in the vicinity of the section is estimated based on the rules stored in the control unit, and the font selection unit is controlled to reset the font type after the font change point.

FIG. 1 is a block diagram showing a schematic functional configuration of a character recognition device according to a first embodiment. FIG. 2 is a schematic view showing a configuration example of a work data storage area provided in the character recognition device of the first embodiment. 6 is a flowchart showing the procedure of character recognition processing by the character recognition device of the first embodiment; FIG. 7 is a schematic view showing contents written in a work data storage area during processing by the character recognition device of the first embodiment. The schematic diagram for demonstrating the comprehensive score calculated by the selection reconsideration judgment part of 1st Embodiment. The schematic for demonstrating the specific example of the comprehensive score calculated by the selection reconsideration judgment part of 1st Embodiment. FIG. 7 is a schematic diagram for explaining processing in which a selection reconsideration determination unit according to the first embodiment determines whether a font change point candidate is actually a font change point. FIG. 5 is a schematic diagram for explaining processing for identifying and specifying a font by the font selection unit of the first embodiment. FIG. 7 is a schematic view for explaining processing of the font selection unit of the first embodiment for identifying a font. FIG. 8 is a block diagram showing a schematic functional configuration of a character recognition device according to a second embodiment. FIG. 14 is a schematic view for explaining the process of font change point estimation by the selection reconsideration determination unit of the second embodiment;

  Hereinafter, a character recognition device, a character recognition method, and a program according to an embodiment will be described with reference to the drawings.

First Embodiment
FIG. 1 is a block diagram showing a schematic functional configuration of the character recognition device according to the present embodiment. As shown in the figure, the character recognition device 1 selects an image input unit 21, an image processing unit 22, a character extraction unit 23, a recognition database 30, a character recognition unit 31, a font selection unit 33, and It has a reconsideration determination unit 35 and an output unit 41.

  The image input unit 21 acquires an image from the outside. This image includes a string of characters to be subjected to character recognition. For example, the image input unit 21 acquires a two-dimensional image by optically reading a paper document placed on a glass surface or the like. The image input unit 21 may be provided with a feeder for reading and feeding a plurality of sheets of paper sequentially. Further, for example, the image input unit 21 may acquire an image file supplied from the outside. Here, the image file supplied from the outside holds the information of the image in an appropriate format such as PDF format, JPEG format, PNG format, TIFF format and the like. "PDF" is an abbreviation for Portable Document Format. Also, "JPEG" is an abbreviation of Joint Photographic Experts Group. Also, "PNG" is an abbreviation of Portable Network Graphics. Also, "TIFF" is an abbreviation of Tagged Image File Format.

The image processing unit 22 performs an appropriate pre-processing on the image acquired by the image input unit 21 so as to enable character recognition. An example of pre-processing is image binarization. The image processing unit 22 binarizes the image into a ground color (for example, white) and a character color (for example, black). The image processing unit 22 appropriately performs processing such as normalization of the size of the character area, removal of noise on the image, and smoothing of character lines, as needed, in addition to the binarization processing. Note that these processes performed by the image processing unit 22 can be performed using the existing technology.
The character extraction unit 23 extracts individual character portions from the image processed by the image processing unit 22 and separates the characters into individual characters. The image received by the character extraction unit 23 includes a character string, and the character extraction unit 23 cuts out the image of each character contained in the character string, and cuts out the image according to the arrangement order of the characters in the character string Save the image of each character. Note that the character extraction and cutout processing itself can be performed using existing techniques.

  The recognition database 30 holds dictionary data to be used for character recognition. The dictionary data includes information of an image of each character, information of a feature amount for recognition of each character, or both of them. The recognition database 30 holds dictionary data for each font type and each character type. Here, the font type is a type identified by a combination of a font family and a character decoration attribute. The font family is roughly classified into, for example, a Gothic font, a Mincho font, a textbook font, and the like. The font family can be further classified finely according to the font, even among the above roughly classified groups (eg, Gothic type), and also finely classified according to the maker of the font. The character decoration attributes are, for example, italics (represented as diagonal characters) or bold (represented by thick lines), and the font is further identified according to whether or not these attributes are included. It is also good. For example, the recognition database may hold dictionary data corresponding to several to several hundreds of font types.

The character recognition unit 31 recognizes characters extracted by the character extraction unit 23 while referring to the recognition database 30. The character recognition unit 31 reads out dictionary data of the font type selected by the font selection unit 33 described later from the recognition database 30, and uses it for character recognition processing. Specifically, the character recognition unit 31 calculates a score for each character type based on the image data for one character passed from the character extraction unit 23. The score is a numerical value representing the degree of match between the dictionary data and the character image. Then, the character recognition unit 31 outputs the character type with the highest score as the recognition result of the image. The character recognition unit 31 outputs information of the character code (ASCII code, JIS code, etc.) of the character with the highest score as the recognition result. In the process of character recognition processing, a string of pairs of character code and its score, which are arranged in order of certainty, is obtained. In the present embodiment, the score is represented as a value of 1000 full marks.
In addition, the existing technology can be used as processing itself (character recognition) which specifies the character contained in the image data by comparing the input image data and dictionary data.

The font selection unit 33 specifies the font type of the character by comparing the character extracted by the character extraction unit 23 with the dictionary data stored in the recognition database 30. Further, the font selection unit 33 sets a font to be used when the character recognition unit 31 performs recognition processing according to the result of specifying the font type. The font selection unit 33 may reselect a font under the control of the selection reconsideration determination unit 35. That is, there is a case where the font selection unit 33 resets the font while the font for recognition processing is already set.
The details of the process in which the font selection unit 33 selects a font will be described later.

  The selection reconsideration determination unit 35 receives the result from the character recognition unit 31 for each character, and determines from the situation of the score in the recognition processing whether or not the target font has been changed. Specifically, the selection reconsideration determination unit 35 receives the data of the score that is the result of the recognition process from the character recognition unit 31, and the font is changed in the middle of the character string in the recognition process based on the score sequence. To judge. That is, when the character recognition unit 31 sequentially performs character image recognition processing, the selection reconsideration determination unit 35 detects a section in which the score is lowered in the string of character images. Then, the selection reconsideration determination unit 35 estimates the position where the font in the section or the vicinity of the section has been changed based on the rule. Thereby, the selection reconsideration determination unit 35 determines whether or not the font type set in the recognition process should be changed. In addition, when determining that the font has been changed, the selection reconsideration determination unit 35 specifies from which position in the character string the font has been changed. The position where the font has been changed may be called "font change point". If it is determined that the font has been changed, the selection reconsideration determination unit 35 controls the font selection unit 33 to reselect the font.

  The font does not usually change for each character in a printed document or the like, and once selected, the font is fixedly used at least at a certain length. The selection reconsideration determination unit 35 makes a determination based on the characteristics of such a document. The selection reconsideration determination unit 35 internally includes a rule base, and the rule base holds knowledge of the point (position) at which the font is changed based on the above-mentioned characteristics of the document. The selection reconsideration determination unit 35 determines whether font selection should be reconsidered based on the rules (knowledge) stored in the rule base. Details of the process will be described later.

  The output unit 41 outputs the result of the recognition process by the character recognition unit 31 to the outside. The output unit 41 outputs at least a string of character codes which is a recognition result. In addition, the output unit 41 may output the value of the score of the character as the recognition result. Further, the output unit 41 may output the code of the character whose score is relatively high among the characters not selected as the recognition result and the value of the score of each character. The content and format of the data output by the output unit 41 may be determined as appropriate.

  FIG. 2 is a schematic view showing a configuration example of a work data storage area when the character recognition device 1 performs character recognition processing of a character string. The illustrated data is temporarily recorded in a semiconductor memory, a magnetic hard disk drive, or the like during processing. The illustrated data is two-dimensional tabular data, and the horizontal direction corresponds to the position of the character in the character string. This data includes the items of character count, input character image, score, adopted font, and recognition result.

The character count is a count of the number of characters from the beginning of the image (image such as a document) acquired by the image input unit 21. The character count value is an integer starting at one. The range of data shown in the figure includes 99 to 119 as the character count. Each data of the input image, the score, the adopted font and the recognition result is held in association with the character count.
The input character image is data of an image of each character extracted by the character extraction unit 23. This input character image is a target of recognition processing by the character recognition unit 31.

The score is data of a score obtained as a result of recognition processing. As data of the score, not only the score for the character finally determined as the recognition result (the highest score) but also the score for other characters may be held. Furthermore, when scores are calculated using a plurality of font types, values of a plurality of scores may be held in association with a combination of font types and character types.
The adopted font holds data for identifying the type of font selected by the font selection unit 33 for the character. When the font selection unit 33 resets the font selection, the data of the adopted font is also rewritten for the character.
The recognition result holds the type of recognized character. Specifically, the character code is held as the recognition result. If, for example, recognition processing is performed multiple times as a result of font reselection, the recognition result may be rewritten.

  The current point is index information that indicates the position of the character that is currently the target of recognition in the processing in the character recognition device 1. As the character recognition unit 31 performs recognition processing one character at a time, the current point advances one by one (from left to right in the figure). However, there is also a possibility that the current point moves back (to the left in the figure) due to font reselection or the like.

  In the example shown in the figure, the value of the character count is from 97 to 119 in increments of one. Illustration is omitted about the part whose value of a character count is 96 or less and 120 or more. Further, the input character images corresponding to these character counts are images of characters "SHI", "SHI", "SHI", "He", "HA",... (Hereinafter omitted). In addition, the font used for private may be changed on the way. Specific examples of score data are omitted in the figure. The adopted font is represented by the identification data "A" in the character count range of 99 to 108. In addition, the part whose character count is 109 or more is a part after the present point, and since the adopted font is seen, it is shown as "-". The recognition result is a recognition result corresponding to each character image. Here, from the position of the character count 99, "S", "S", "S", "S", "He", "H", "," "E", "Ear", "Sunny", "S", "Re" The recognition result "" is stored. The recognition result may include recognition errors.

  FIG. 3 is a flowchart showing the procedure of the character recognition process by the character recognition device 1. The overall flow of processing by the character recognition device 1 will be described below along the flowchart.

  First, in step S1, the character recognition device 1 initializes the current point. Specifically, the character recognition device 1 sets the current point to the beginning of the character string to be recognized.

Next, in step S2, the font selection unit 33 selects and sets a font. Specifically, when the process proceeds from step S1 to step S2, the font selection unit 33 selects and sets a predetermined default font. When the process proceeds from step S9 to step S2, the font selection unit 33 executes the process of step S8 (described later) based on the image of characters after the font change position (the font changed position will be described later). Select and set the font identified as a result of.
Even when the process proceeds from step S1 to step S2, the font selection unit 33 may identify and select the font based on the image of the characters of several characters after the current point.
The details of the process in which the font selection unit 33 selects a font will be described later.

  Next, in step S3, the character recognition unit 31 performs recognition processing of the character of the current point. Specifically, the character recognition unit 31 uses the dictionary data of the currently selected font to calculate the score for each character of the character image of the current point. Then, the character with the highest score is determined as the recognition result. At this time, the character recognition unit 31 writes the information of the calculated score in the data storage area for work (see FIG. 2).

  Next, in step S4, the selection reconsideration determination unit 35 calculates an overall score for the character string of the predetermined length based on the score of the character string of the predetermined length up to the current point. Here, the "predetermined length" may be plural. The total score calculated here is a numerical value for evaluating whether or not the currently selected font is a valid font for the character string of the predetermined length. If the total score is equal to or higher than a predetermined reference value, it can be estimated that the currently selected font is a valid font. Conversely, if this total score is less than the reference value, it can be estimated that the currently selected font may not be appropriate. In addition, there are a calculation method based on an absolute score and a calculation method based on a relative score by comparison with the immediately preceding score. The method of calculating these total scores will be described in detail later.

  Next, in step S5, the selection reconsideration determination unit 35 determines whether the total score calculated in step S4 is equal to or higher than a reference value. As described above, when the “predetermined length” is more than one, more than one total score is also calculated. If all the calculated total scores are equal to or higher than the reference value, that is, if it is estimated that the currently selected font is valid (step S5: YES), the process proceeds to step S10. If one or more total scores are less than the reference value, that is, if the currently selected font may be reconsidered (step S5: NO), the process proceeds to step S6.

Next, when the process proceeds to step S6, in the same step, the selection reconsideration determination unit 35 detects the position at which the font has been changed. At this time, the selection reconsideration determination unit 35 detects the font change position based on the knowledge. The knowledge in the present embodiment is that "the font change position is limited to the position of the word break". In other words, it is the knowledge that the font is not changed in the middle of the word. Therefore, the selection reconsideration determination unit 35 internally includes word dictionary data, and collates with the data storage area for work while referring to the word dictionary data to detect a font change position.
The selection reconsideration determination unit 35 may include a morphological analysis processing engine. In this case, the selection reconsideration determination unit 35 performs morphological analysis processing on the character string up to the current point (the character string of the recognition result). Then, as a result of the morpheme analysis process, the position of the morpheme delimiter in the character string to be recognized is treated as the position of the word delimiter. In addition, even if it considers that a morpheme and a word are substantially the same, it does not interfere in this embodiment.
The process of detecting the font change position will be described in more detail later.
Thus, the selection reconsideration determination unit 35 detects (estimates) the font change position based on the score of the series of characters calculated by the character recognition unit 31 and the knowledge.

  Next, in step S7, the selection reconsideration determination unit 35 determines whether or not the font change position has been detected in the process of step S6. If the font change position has been detected, the process branches here to identify the changed font. If the font change position can be detected (step S7: YES), the process proceeds to the next step S8. If the font change position can not be detected (step S7: NO), the process jumps to step S10.

  Next, when the process proceeds to step S8, in the step, the font selection unit 33 identifies the used font based on the string of character images after the font change position. Specifically, the font selection unit 33 calculates a score using dictionary data of each font for a string of character images after the font change selection position and up to the current point. Then, for the column, it is determined that the font with the highest score is used in the column.

  Next, in step S9, the character recognition device 1 sets the current point as the font change position. As a result, character recognition processing is performed again using the dictionary data of the font identified in step S8 from the font change position detected in step S6. After the process of this step is completed, the process returns to step S2.

Processing from step S5 or from step S7 to step S10 is as described below.
In step S10, the character recognition device 1 advances the current point to the next character.
Next, in step S11, the character recognition device 1 determines whether or not all the input character images to be subjected to the character recognition process have ended. If it has ended (step S11: YES), the character recognition device 1 ends the processing of the entire flowchart. If it has not ended, that is, if an unprocessed input character image remains (step S11: NO), the process returns to step S3 to process the next character.
This is the end of the description of the overall processing procedure of the character recognition device 1.

  Next, the details of the process of selection reconsideration will be described. As described above, the selection reconsideration determination is a process of determining whether or not to reselect the font and estimating a font change point.

  FIG. 4 is a schematic view showing the contents written in the data storage area for work in a state in the middle of processing by the character recognition device 1. The configuration itself of the data storage area for work is as described above. In the figure, the current point indicates the position of the character count "108". The currently selected font is "A", and at each position from the character count "99" to "108", the score calculated with the dictionary data of font "A" is written. Specifically, in the character count "99", the score is 950, and thereafter, the score at each position is 933, 948, 923, 910, 901, 777, 791, 760, 750. The score has not been calculated at positions after the character count “109”. Within the range shown in the figure, in the character count "99" to "104", the value of the score is 900 or more out of 1000 points, respectively. On the other hand, the value of the score is less than 800 points in each of the character count "105" to "108". As described above, in a partial string in a character string to be recognized, the score value may change with discontinuity, which may occur when the font used in the input character image changes. In the example shown in the figure, the score is low after the position of the character count "105". At this time, the selection reconsideration determination unit 35 determines whether or not the font has changed after the position of the character count “105” in the process described below. Further, the selection reconsideration determination unit 35 estimates a point (position) at which the font has changed. As described above, the selection reconsideration determination unit 35 calculates the total score of the predetermined section in the character string for the determination. This is the total score calculated in step S4 in FIG.

FIG. 5 is a schematic diagram for explaining the total score calculated by the selection reconsideration determination unit 35. As shown in FIG. In the figure, each circle corresponds to a character to be recognized. And in the same figure, the position of the character is shown by relative value. That is, the position of the current point is zero, the front (left side in the figure) of the character string is negative, and the rear (right side in the figure) of the character string is negative. Also, the score at each position is represented using this relative value. That is, the score at a relative position "0" is S _0, the score at a relative position "-1" is S _-1, The same applies to the other positions.

Here, the selection reconsideration determination unit 35 calculates the total score of the string of the latest n characters including the current point. The n characters are characters having a relative position of “−n + 1” to “0” in the same figure. However, the range which n can take is N _min ≦ n ≦ N _max . N _min and N _max are integer values appropriately determined. For example, the values of N _min and N _max can be determined as setting values of the character recognition device 1. As an example, N _min = 1 and N _max = 6 but they are not limited to these values. The selection reconsideration determination unit 35 calculates the total score of the above-described n character string based on the score values S _{−n + 1} , S _{−n +2} ,..., S ₀ of the respective character scores. The higher the score value of each character included in the n, the higher the value of the n overall scores.

As the overall score calculated by the selection reconsideration determination unit 35, either an absolute overall score or a relative overall score is used. About which of these is used, it determines suitably.
The absolute total score is an absolute value determined only by the score of the n characters. As an example, the selection reconsideration determination unit 35 uses an average value of these n scores as an absolute total score. That is, at this time, the absolute comprehensive score is
(S- _{n + 1} + S- _{n + 2} + ... + S ₀ ) / n
It is calculated by the formula
The relative total score is a relative value indicating how much the score of the n characters is higher or lower than the score of the string of characters of a predetermined length immediately preceding it. As an example, the selection reconsideration determination unit 35 uses a value obtained by subtracting the average value of the N ₀ scores immediately before the string of n characters from the average value of the n scores. Note that N ₀ is an integer value that is appropriately determined. As an example, N ₀ = N _max may be used, but is not limited to that value. That is, at this time, the relative comprehensive score is
_{{(S -n + 1 + S} -n + 2 + ··· + S 0) / n} - {(S -n-N0 + 1 + S -n-N0 + 2 + ··· + S -n)} / N 0}
It is calculated by the formula

When n can take a plurality of values, the selection reconsideration determination unit 35 calculates a total score for each value of n. In the case of N _min = 1, N _max = 6 exemplified above, n can take six values: n = 1, 2, 3, 4, 5, 6.

FIG. 6 is a schematic diagram for describing a specific example of the comprehensive score described above. In the example shown in the figure, n = 4 and N ₀ = 4. The values 948, 923, 910, 901, 777, 791, 760, 750 shown in the figure are scores calculated for each position.
In the illustrated example, the selection reconsideration determination unit 35 calculates the absolute total score as follows. That is,
_{(S -3 + S -2 + S} -1 + S 0) / n
= (777 + 791 + 760 + 750) / 4
= 769.50
It is.
Further, in the illustrated example, the selection reconsideration determination unit 35 calculates the relative total score as follows. That is,
_{{(S -3 + S -2 +} S -1 + S 0) / n} - {(S -7 + S -6 + S -5 + S -4)} / N 0}
= {(777 + 791 + 760 + 750) / 4}-{(948 + 923 + 910 + 901) / 4}
= -151.00
It is.

  As described above, when the total score (absolute total score or relative total score) calculated by the selection reconsideration determination unit 35 is less than a predetermined reference value, the position of the head of the n character string is , And may be a font change point (branch in step S5 of FIG. 3). That is, the position is a font change point candidate. When n can take a plurality of values, there may be a plurality of font change point candidates.

Here, the reference value for the overall score can be set as appropriate.
The reference value for the absolute total score is, for example, 790 points.
The reference value for the relative overall score is, for example, -100.
In addition, you may use as a reference value except the value illustrated here.

  FIG. 7 is a schematic diagram for describing processing in which the selection reconsideration determination unit 35 determines whether or not each of the font change point candidates is actually a font change point. As described above, in order to estimate the font change point, the selection reconsideration determination unit 35 performs processing based on the knowledge that “the font is changed at the word break”. The selection reconsideration determination unit 35 internally includes means for storing word dictionary data, and estimates a font change point while referring to the word dictionary data. Each of (a) and (b) of the same figure shows the example of the font change point candidate which the selection reconsideration judgment part 35 calculated | required based on the above-mentioned total score. The figure (a) shows an example in case a font change point is "element" in "it is wonderful." At this time, there is a possibility that different fonts are used between this "element" and the immediately preceding "," (dot). The figure (b) shows an example in case a font change point is "it" in "it is wonderful." At this time, it is possible that different fonts are used between this “shi” and the immediately preceding “ra”.

  The selection reconsideration determination unit 35 performs processing with reference to the word dictionary data for each of the font change point candidates. The word dictionary data is prepared in advance, and includes, for example, words such as "but", "he", "ha", "wonderful", "good" and the like. Therefore, since the font change point candidate in the example of FIG. 6A is at the delimitation of the word “ha” and the word “superb”, the selection reconsideration determination unit 35 determines that the font change point candidate is a true font. It determines that it is a change point. On the other hand, since the font change point candidate in the example (b) of the selection reconsideration determination unit 35 is located at the position where the character string “superb” is separated from the character string “i” The candidate is determined not to be a true font change point. As its premise, the character string "Superb" and the character string "I" are not registered in the word dictionary data.

  When the selection reconsideration determination unit 35 includes a morpheme analysis processing engine, the selection reconsideration determination unit 35 performs morpheme analysis processing on the target character string. Then, if the font change point candidate is located at the morpheme delimiter, the selection reconsideration determination unit 35 determines that the font change point candidate is a true font change point. If the font change point candidate is not a morpheme delimiter, the selection reconsideration determination unit 35 determines that the font change point candidate is not a true font change point. The morphological analysis processing itself can be performed using an existing technology.

Next, details of processing by the font selection unit 33 will be described.
The font selection unit 33 identifies the font type used in the character image based on a string of character images of a predetermined length. The font selection unit 33 performs a process of selecting a font in step S2 of FIG.

  FIG. 8 is a schematic diagram for explaining a process in which the font selection unit 33 identifies and identifies a font. The row of "but he" character images shown in FIG. 6A is a character image which is a target of the font selection process performed by the font selection unit 33. The font selection unit 33 first specifies a font for each of these character images. In the illustrated example, the font of the leading "" is first identified. Therefore, the font selection unit 33 selects a plurality of candidate characters related to this character image. The figure (b) shows the character candidate selected, and the character candidate includes "hi" in hiragana, "re" in katakana, and "L" in alphabetic (upper case) letters. In order to select these character candidates, the font selection unit 33 uses a dictionary for candidate character specification. The candidate character specification dictionary is stored in the recognition database 30, and is dictionary data having a feature amount common to each font. That is, the candidate character specification dictionary is dictionary data that does not depend on the font type. The candidate character specification dictionary contains information such as the common feature amount of each character in a plurality of font types, and is written in the recognition database 30 in advance. At this stage, since the font has not been identified, the font selection unit 33 identifies a plurality of candidate characters by using this candidate character identification dictionary. By using the common feature amount for the font type, it is possible to narrow down the candidates from the character shape obtained from the target image. Moreover, it is also possible to include the similar character in the set of candidate characters by registering the similar character of the acquired character candidate in the candidate character specifying dictionary in advance.

  When the candidate characters shown in FIG. 6B are obtained, the font selection unit 33 then refers to dictionary data for each font, and calculates the score of the character to be processed. The figure (c) shows the example of the score computed for every candidate character and for every font classification about the character image of a process target. In the table of FIG. 6C, each row corresponds to a font type, and each column corresponds to a character type (candidate character). Among the score values shown in this table, the highest is the score 960 of the character "B" in the font type "font A". Therefore, the font selection unit 33 specifies that the font of this character image is "font A". Also, the font selection unit 33 performs the same processing on other characters that are targets of font identification, and specifies the font of each character.

  FIG. 9 is a schematic diagram for explaining the process of the font selection unit 33 identifying a font, and shows the result of font identification performed on a plurality of character images. As shown in the figure, the identification results for each character image of "S", "S", "S", "He" and "H" from the left side are "Font A", "Font B" and "Font" respectively. A "," font A ", and" font A ". In this manner, the font selection unit 33 repeats the process of specifying the font over several characters in order to increase the accuracy of font selection, and selects the most probable font. The font selection unit 33 specifies the entire font by, for example, a simple majority rule, based on the font specified for each of the plurality of characters. In the illustrated example, "font A" is selected.

  At this time, a weight (score for font selection) may be set in advance for each character type, and the entire font may be specified by a weighted majority. As the font selection score, a lower value (a smaller weight) is set for a character having a simple glyph (for example, a symbol such as an addition symbol “+” or a colon “:”). Conversely, a high value (a value with a large weight) is set for a character having a complicated character shape (for example, kanji or the like).

  It is also possible to select a font with a high score using dictionary data for each font type from the beginning without narrowing and extracting candidate characters by the candidate character specifying dictionary described in FIG. In addition, the font selection unit 33 may select a font by a similar method.

As described above, in the first embodiment, the selection reconsideration determination unit 35 detects the possibility of the font change in the middle based on the data of the score output from the character recognition unit 31, and Estimate the position where the font was changed based on it. Thereby, the selection reconsideration determination unit 35 determines whether or not the font type set in the recognition process should be changed. Therefore, in the first embodiment, the position where the font is changed can be detected with high accuracy.
In the first embodiment, the selection reconsideration determination unit 35 estimates the position where the font has been changed, based on the rule (knowledge) that only the word break can be the font break. This makes it possible to accurately detect the position at which the font has been changed without making a determination that the font has been changed in the middle of the word.
Then, the accuracy of character recognition is improved by these.

Second Embodiment
Next, a second embodiment will be described. In addition, description may be abbreviate | omitted about the matter similar to the above-mentioned embodiment. In the following, matters specific to the present embodiment will be mainly described.
FIG. 10 is a block diagram showing a schematic functional configuration of the character recognition device according to the present embodiment. As shown in the figure, the character recognition device 2 selects an image input unit 21, an image processing unit 22, a character extraction unit 23, a recognition database 30, a character recognition unit 31, a font selection unit 33, and A reconsideration determination unit 36 and an output unit 41 are provided. That is, the character recognition device 2 has a selection reconsideration determination unit 36 in place of the selection reconsideration determination unit 35 in the first embodiment.

The entire processing procedure by the character recognition device 2 is substantially the same as that of the character recognition device 1 described with reference to FIG. However, the knowledge used in step S6 in the same figure is different from the knowledge used in the first embodiment. The process of the selection reconsideration determination unit 36 will be described below. The selection reconsideration determination unit 36 in the present embodiment uses the following two rules as knowledge. These rules are stored in the rule base already described in the first embodiment.
(Rule 1) A series of characters separated by a pair of parentheses may be printed in a specific font.
(Rule 2) A series of numeric strings beginning with a currency symbol (but may contain commas or periods) may be printed in a specific font.

Examples of “brackets and the like” in the above-mentioned rule 1 include brackets (single or double brackets), parentheses, brackets, curly brackets (curly braces), angle brackets, etc. It is not limited. These parentheses are used in the document in left and right pairs (but in the case of vertical writing, upper and lower pairs). These pairs of parentheses and the like are called “enclosed symbols” for convenience. An enclosure symbol is a symbol for enclosing a character string from the left and right (but from top and bottom in the case of a vertical writing document). The symbol located at the beginning of the box is also called "open parenthesis". Also, the symbol located at the end of the box is also referred to as "closed parenthesis". In other words, rule 1 is a rule that, regarding a character string surrounded by a pair of enclosure symbols in the input string, the font type is changed between the character string before and after the character string.
The “currency symbol” in rule 2 above includes the symbol “¥” representing a yen, the symbol “$” representing a dollar, and symbols representing other currencies. Note that the currency symbol and the sequence of numbers following the currency symbol are collectively referred to as “number sequence” for convenience. That is, rule 2 is a rule that the type of the numeral string in the string is changed between the character string before and the character string after the numeral string.

  There are cases where so-called full-size characters are used as the above-mentioned parentheses or the like or currency symbols, and cases where so-called half-size characters are used. Regardless of whether these characters are full-width or half-width, the selection reconsideration determination unit 36 infers font change points while applying Rule 1 or Rule 2.

  FIG. 11 is a schematic diagram for explaining the process of font change point estimation by the selection reconsideration determination unit 36. As shown in FIG. Each of the figures (a) to (c) shows a string of character images to be recognized. In addition, in the same figure, while showing the present point in recognition processing, the relative position on the basis of the present point is attached on each character image.

  In the example of (a) of the figure, the character of the current point (relative position is “0”) is “i”, the character at relative position “-5” is the left double bracket, relative position “+4” The characters in are right double brackets. Note that the total points of the partial character string based on the current position have already been calculated (the calculation of the total points is the same as in the first embodiment). Thus, the selection reconsideration determination unit 36 has already obtained font change point candidates. If the character at the relative position "-5" is a font change point candidate, the selection reconsideration determination unit 36 determines that the font change point candidate is a true font change point according to the above-described rule 1. That is, it is determined that different fonts are used for the relative position "-6" and the relative position "-5". On the other hand, when the character at the relative position "-2" is a font change point candidate, according to rule 1 described above, the selection reconsideration determination unit 36 determines that this font change point candidate is not a true font change point. . That is, it is determined using the knowledge (rule 1) that different fonts are not used for the relative position "-3" and the relative position "-2".

  In the example of (b) of the figure, the character of the current point (relative position is “0”) is “be”, the character at relative position “-12” is the left double bracket, relative position “+3” The characters in are right double brackets. Here again, the selection reconsideration determination unit 36 has already obtained font change point candidates. If the character at the relative position "-5" is a font change point candidate, according to rule 1 described above, the selection reconsideration determination unit 36 determines that the font change point candidate is not a true font change point. That is, the determination is performed using the knowledge that different fonts are not used for the relative position "-6" and the relative position "-5". On the other hand, if the character at the relative position "-2" is a font change point candidate, the selection reconsideration determination unit 36 determines that the font change point candidate is a true font change point according to the above-described rule 1. That is, it is determined that different fonts are used for the relative position "-3" and the relative position "-2".

  In the example of (c) of the figure, the character of the current point (the relative position is "0") is the numeral "5", and the character at the relative position "-5" is the currency symbol "$". In addition, a string of characters from the relative position "-5" to "0" is a series of digit strings starting with a currency symbol. If the character at the relative position "-5" is a font change point candidate, the selection reconsideration determination unit 36 determines that the font change point candidate is a true font change point according to the above-described rule 2. That is, it is determined that different fonts are used for the relative position "-6" and the relative position "-5". On the other hand, when the character at the relative position "-3" is a font change point candidate, according to the above-mentioned rule 2, the selection reconsideration determination unit 36 determines that this font change point candidate is not a true font change point. Do. The reason is that the relative position “-3” is in the middle of “a series of digit strings starting with a currency symbol” defined in rule 2. That is, the selection reconsideration determination unit 36 performs the determination using the knowledge that the font is not changed in the middle of such a series.

  As described above, the selection reconsideration determination unit 36 uses the rule 1 and the rule 2 in combination to estimate the font change point. Note that the selection reconsideration determination unit 36 may use only one of the rule 1 and the rule 2. In addition, other rules may be stored in advance in the rule base, and the selection reconsideration determination unit 36 may estimate the font change point using the other rules in combination. In addition, the rule (knowledge) that “the font change position is the position of the word break” described in the first embodiment by the selection reconsideration determination unit 36 and the rule described in the present embodiment are used in combination. You may do it.

As described above, in the second embodiment, the font of the character string enclosed by the surrounding symbols (a pair of left and right parentheses, etc.) of the selection reconsideration determination unit 36 is a font different from the character string before and after it ( Estimate the position where the font has been changed based on the following rule. This improves the accuracy of estimating font change points.
Further, in the second embodiment, the selection reconsideration determination unit 36 determines a number string (a string consisting of only numbers, a string consisting only of numbers, periods and commas, a symbol having a currency symbol at the beginning of these strings, etc.) The position of the font is estimated based on the rule that the font of (in some cases) is a font different from the preceding and following character strings. This improves the accuracy of estimating font change points.
Furthermore, in the second embodiment, it is also possible to use the knowledge that only the position of the word break can be a font change point described in the first embodiment. This improves the accuracy of estimating font change points.

  In each of the above embodiments, a plurality of pieces of knowledge (rules) used when the selection reconsideration determination unit estimates a font change point are described. However, the following rule may be used. For example, regardless of the presence or absence of a currency symbol, a rule may be used in which a series of numeric strings precede and follow is a break, and the font is changed at the place of the break. Also, for example, a rule may be used in which a series of alphabetic strings in a Japanese sentence is preceded and followed by a break, and the font is changed at the place of the break.

  In each of the above embodiments, according to the flowchart shown in FIG. 3, the case where control is sequentially passed to the character recognition unit, the selection reconsideration determination unit, and the font selection unit and processing is sequentially performed is described. The functions of the recognition unit, the selection reconsideration determination unit, and the font selection unit may be allocated to a plurality of processors, and at least a part of the functions may be processed in parallel. This makes it possible to execute character recognition processing faster while selecting an appropriate font.

  In each of the above embodiments, the processing of the character recognition unit, the processing of the selection reconsideration judgment unit, and the processing of the font selection unit are performed in one pass for one input image. The pass of the process of identification and the pass of the process of character recognition based on the font specified in the pass may be executed by a two-pass method.

  Further, the function of the image processing unit in each of the above embodiments may be provided outside the character recognition apparatus. In this case, the character recognition device inputs the preprocessed image. Furthermore, the function of the character extraction unit in each of the above embodiments may be provided outside the character recognition device. In this case, the character recognition device inputs an ordered sequence of clipped character images.

  According to at least one embodiment described above, the possibility of font change on the way is detected based on the data of the score output from the character recognition unit, and the font is also detected based on the rule stored in the rule base The position (font change point) at which the font has been changed is determined by having a selection reconsideration judgment unit that estimates the position at which the character has been changed and determines whether or not the font type set in the recognition process should be changed. It can detect accurately. This leads to an improvement in the accuracy of character recognition.

  Note that at least a part of the functions of the character recognition device in each embodiment described above may be realized by a computer. In that case, the program for realizing the function may be recorded in a computer readable recording medium, and the program recorded in the recording medium may be read and executed by a computer system. Here, the “computer system” includes an OS and hardware such as peripheral devices. The term "computer-readable recording medium" refers to a storage medium such as a flexible disk, a magneto-optical disk, a ROM, a portable medium such as a ROM or a CD-ROM, or a hard disk incorporated in a computer system. Furthermore, "computer-readable recording medium" holds a program dynamically for a short time, like a communication line in the case of transmitting a program via a network such as the Internet or a communication line such as a telephone line. It may also include one that holds a program for a certain period of time, such as volatile memory in a computer system that becomes a server or client in that case. The program may be for realizing a part of the functions described above, or may be realized in combination with the program already recorded in the computer system.

  While certain embodiments of the present invention have been described, these embodiments have been presented by way of example only, and are not intended to limit the scope of the invention. These embodiments can be implemented in other various forms, and various omissions, replacements, and modifications can be made without departing from the scope of the invention. These embodiments and modifications thereof are included in the invention described in the claims and the equivalents thereof as well as included in the scope and the gist of the invention.

1, 2 ... character recognition device, 21 ... image input unit, 22 ... image processing unit, 23 ... character extraction unit, 30 ... recognition database, 31 ... character recognition unit, 33 ... font selection unit, 35, 36 ... selection reconsideration Judgment part, 41 ... output part

Claims (6)

A recognition database that holds dictionary data for character recognition for each font type;
While acquiring a string of character images and performing recognition processing of the character image while referring to the dictionary data in the recognition database, the matching degree between the dictionary data and the character image in the recognition processing is obtained. A character recognition unit that outputs a score representing
A font for setting a font type of the dictionary data to be referred to by the character recognition unit by specifying a font type of the character image based on the character image and the dictionary data for each font type in the recognition database A selection unit,
When the character recognition unit sequentially performs recognition processing of the character image included in the string of the character image, a section in the string whose score has been lowered is detected, and a rule stored in the rule base A selection reconsideration determination unit that estimates a font change point in the section or in the vicinity of the section based on, and controls the font selection section to reset the font type after the font change point;
Character recognition apparatus comprising: The selection reconsideration determination unit estimates the font change point based on the rule that the font type is changed only at word break positions in the column.
The character recognition device according to claim 1. The selection reconsideration determination unit is based on the rule that a character string surrounded by a pair of enclosure symbols in the string is changed in font type with a character string before and after the character string. To estimate the font change point,
The character recognition device according to claim 1 or 2. The selection reconsideration determination unit estimates the font change point based on the rule that a digit string in the string is a character string before and after a character string of the digit string is changed in font type. Do,
The character recognition device according to any one of claims 1 to 3. The character image recognition process is performed while acquiring a string of character images and referring to the dictionary data in a recognition database that holds dictionary data for character recognition for each font type, and performing the character image recognition processing, and the dictionary at the time of recognition processing A character recognition process that outputs a score representing the degree of match between the data and the character image;
A font for setting a font type of the dictionary data to be referred to in the character recognition process by specifying a font type of the character image based on the character image and the dictionary data for each font type in the recognition database The selection process,
When recognition processing of the character image included in the string of the character image is sequentially performed in the character recognition process, a section in the string whose score is lowered is detected, and a rule stored in a rule base A selection reconsideration determining step of estimating a font change point in the section or in the vicinity of the section based on and selecting the font selection process to reset the font type after the font change point;
Character recognition method including. A recognition database holding dictionary data for character recognition for each font type,
While acquiring a string of character images and performing recognition processing of the character image while referring to the dictionary data in the recognition database, the matching degree between the dictionary data and the character image in the recognition processing is obtained. A character recognition unit that outputs a score representing
A font for setting a font type of the dictionary data to be referred to by the character recognition unit by specifying a font type of the character image based on the character image and the dictionary data for each font type in the recognition database A selection unit,
When the character recognition unit sequentially performs recognition processing of the character image included in the string of the character image, a section in the string whose score has been lowered is detected, and a rule stored in the rule base A selection reconsideration determination unit that estimates a font change point in the section or in the vicinity of the section on the basis of, and controls the font selection section to reset the font type after the font change point;
Program to function as.

Images