JP3930466B2 - Character recognition device, character recognition program - Google Patents

Description

  The present invention relates to a character recognition apparatus and a character recognition program for reflecting a result of correction on a character recognition result in character recognition processing after correction.

  Conventionally, in a character recognition device, a character recognition result output by character recognition can be corrected by an operator, and the correction result can be reflected in subsequent character recognition processing.

  Various methods have been proposed to reflect the correction information, but once the recognition result has been re-entered (character) pattern that is very similar in image to the corrected (character) pattern. Even in such a case, the correction by the operator may not be reflected in the recognition result. In such a case, the operator had to make corrections again, which required very complicated work.

  For example, in the method described in Patent Document 1, correction information for correcting an evaluation value for each character category is obtained and registered so that a correct character string selected by the operator becomes a recognition result, and subsequent character recognition is performed. By correcting the category evaluation value, correction information by the operator is reflected in the recognition result.

  However, in this method, since the unit of correction of the evaluation value is a category unit, a pattern that is easily confused with an incorrect category in the recognition process is adversely affected by the correction value of the category, and the correction specific to the pattern is not included in the subsequent recognition results. It is hard to be reflected.

  Further, in the method of Patent Document 2, the correction amount of the distance value is obtained so that the correct reading category becomes the first candidate for the pair of the misreading category and the correct reading category corrected by the operator, and the correction amount and the misreading are calculated. An entry composed of a category and the correct reading category is registered in the correction table. If the first candidate matches the misread category in subsequent character recognition, the correct reading category distance value stored in the same entry as the previous misread category is corrected by the correction amount recorded in the entry. Thus, the correct reading category input by the operator in the previous correction is easily output as the recognition result.

  However, even in this method, since the correction amount of the distance value of the correct reading category is registered in units of misreading categories, the category that is difficult to separate in the recognition process is still difficult to separate even if correction is performed, and pattern-specific correction is performed. Even if it is performed by the operator, it is difficult to reflect it in the recognition result. As a result, the modification affects even patterns that are somewhat similar but not so similar.

  Further, the method of Patent Document 3 includes a plurality of recognition engines, and each recognition engine outputs a recognition result reliability coefficient together with the recognition result. As the recognition result of the entire recognition device, the recognition result of the recognition engine having the maximum product of the reliability coefficient of the recognition result for each recognition engine and the reliability coefficient obtained for the recognition engine itself is selected. The number of misreads for each recognition engine is counted using the recognition result corrected by the operator, and the reliability coefficient of the recognition engine is obtained based on the number of misreads. In this method, correction information by the operator is reflected in the recognition result via a reliability coefficient for each engine.

  However, in this method, since the unit of adjustment of the reliability coefficient is the entire recognition target region, the correction information is not selectively reflected in a pattern that is very similar in image to the corrected pattern.

  Also, among character recognition methods that use a recognition dictionary to recognize individual characters, a method is known in which a combination of a corrected recognition result and an input pattern is learned online, and the recognition dictionary is sequentially corrected. ing. For example, in the method of Patent Document 4, data captured by a scanner is cut out for each character and processed into a small amount of data (acquired character pattern) from which a feature amount is extracted. When the character recognition result for the acquired character pattern is misread, it can be changed to the character code input from the keyboard. The acquired character pattern to be changed and the changed character code are associated with each other and stored in the recognition dictionary.

In this method, the correction by the operator is reflected through a recognition dictionary and character recognition means using the recognition dictionary. However, since the general character recognition means requires generalization ability, in learning the dictionary, each learning pattern (acquired character pattern representing the feature amount of the character) is not only a pattern very similar to the learning pattern, but in some cases This is performed so as to affect the recognition result of a pattern that is not very similar to the learning pattern. Therefore, in some cases, correction may be made so that a pattern that has been correctly read turns into a misread.
Japanese Patent Laid-Open No. 2002-334302 JP-A-9-237322 Japanese Patent Application Laid-Open No. 06-78561 JP-A-11-143993

  As described above, in the conventional character recognition device, even when a pattern that is very similar to the pattern corrected in the past is inputted, the correction is not performed, or conversely, the pattern that is not very similar to the corrected pattern is also used. There were cases where it had an adverse effect.

  The present invention has been made in consideration of the above-described circumstances. In a pattern that is very similar in image, correction by an operator is surely reflected, and conversely, a pattern that is not very similar is not affected. Thus, a character recognition device and a character recognition program capable of accurately reflecting the correction of the recognition result by the operator in the recognition processing after the correction and obtaining a correct recognition result without increasing the burden on the operator are provided. For the purpose.

The present invention identifies a character recognition means for recognizing a first character image and outputting a first character as a recognition result, and specifying a second character image and a second character corresponding to the second character image A correction dictionary in which an entry including a pair of data to be recorded is superimposed on the first character image and the second character image, and the first character image and the second character image are compared by comparing pixel values of respective pixels. When it is determined that the first character image and the second character image are very similar by the determination unit that determines whether or not the second character image is very similar, the first character correction means for correcting the first character to the recognition result output by said character recognition means, said second character image and the second letter indicating the data for specifying the characters to be paired And before being corrected by the first character correcting means. And presenting means for presenting the second character as the target of manual correction, input means for inputting the third letter for modifying said second character presented by the presenting means by manual correction operation, the presentation a second character correction means for correcting the presented the second character, the third characters input by said input means by means,
When the data for specifying the third character corrected by the second character correcting means is not recorded in the correction dictionary, the first character image and the data for specifying the third character are And a modified dictionary managing means for adding the entry to the modified dictionary .

According to the present invention, it is determined whether the character image to be recognized and the character image recorded in the correction dictionary are very similar, and if it is determined that they are very similar, the character recognition result is corrected. Since the correction is performed by the characters indicated by the data recorded in the dictionary, the pattern that is very similar to the image surely reflects the correction by the operator, and conversely, the pattern that is not very similar is not affected. It is possible to accurately reflect the correction of the recognition result by the operator in the recognition processing after the correction, and to obtain a correct recognition result without increasing the burden on the operator.

  Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing the configuration of a system that implements a character recognition device according to this embodiment.

The character recognition apparatus according to the present embodiment can be configured by incorporating a program for realizing the present embodiment into a computer, and will be described below assuming such a configuration. However, in the present embodiment, each unit can be configured as dedicated hardware, an aggregate thereof, or a computer network for distributed processing. If the configuration includes the units described below, the configuration described here Regardless of what you implement it.

  As shown in FIG. 1, the character recognition device in this embodiment includes a CPU 10, a memory 12, a display 14, a keyboard 15, a pointing device 16, a scanner 17, a storage device 18, and a printer 19.

  The CPU 10 controls the entire character recognition apparatus. According to various programs recorded in the memory 12, the CPU 10 performs data input / output control, control of character recognition processing for characters recorded in a form or the like input from the scanner 17, and the like. Control various processes.

  The memory 12 records programs, data, and the like, and when processing is executed according to various programs, data to be processed is temporarily stored as the programs are executed. As a program stored in the memory 12, for example, a character recognition program 12a for recognizing characters entered in a form or the like is stored. The data stored with the execution of the character recognition program 12a includes image data (original image data 12b) input by the scanner 17 and recognition obtained by character recognition processing on the original image data 12b. Regarding the result, correction dictionary data 12c for correcting the result based on the character image to be recognized, character recognition result data 12d that is a result (character code) of the character recognition process, and the like are stored.

  The display 14 is composed of an LCD (Liquid Crystal Display) or the like, and displays information indicating the character recognition result and the state of the apparatus.

  The keyboard 15 is used for inputting various instructions to the apparatus.

  The pointing device 16 is configured by a mouse, a tablet, or the like, and is used for inputting various instructions to the apparatus.

  The scanner 17 reads an image such as a form to be subjected to character recognition processing and generates original image data.

  The storage device 18 reads / writes programs and data from / to various recording media. The storage device 18 can use a magnetic disk such as a hard disk, an optical disk, a semiconductor memory, or the like as a recording medium. The recording medium may be either a fixed type or a portable type.

  The printer 19 prints a character recognition result for an image such as a form on a paper medium. For example, a result of character recognition processing (character code) for a character entered in an entry frame provided in the form is displayed as an image (character ).

(First embodiment)
Hereinafter, a first embodiment of the present invention will be described with reference to the drawings.
FIG. 2 is a block diagram showing a functional configuration of the character recognition apparatus in the first embodiment realized by the system shown in FIG.

  The character recognition apparatus according to the first embodiment performs preprocessing such as noise removal, binarization, form and preprint character removal, and recognition target area cutout on an input image (original image data 12b). 21, character extraction means 22 for extracting a character group from the image obtained from the preprocessing result, character recognition means 23 for recognizing a character in the extracted character group and converting it into a character code, and recognition by the character recognition means 23 Based on the automatic correction means 24 for automatically correcting the result based on the correction dictionary 25, the manual correction means 26 for enabling the operator to recorrect the correction result by the automatic correction means 24, and the correction in the manual correction means 26 In this way, the registration information (entries) of the correction dictionary 25 is managed, that is, the correction dictionary management means 27 for adding, deleting and changing entries.

  Next, the operation of the character recognition device in the first embodiment will be described with reference to the flowcharts shown in FIGS.

  First, reading is performed on a form or the like to be processed in which characters are entered by the scanner 17, and the original image data 12b (original image 20) is stored in the memory 12 (step A1).

  The pre-processing means 21 performs image processing (pre-processing) such as image segmentation, binarization, noise removal, and the like on the input original image 20 from the periphery of the column to be processed (step A2). ). When the recognition target is a character on the form, image processing such as form removal such as ruled lines and pre-printed characters, and image cutout of a region to be recognized is also performed.

The character extraction unit 22 extracts a character group to be recognized from the image obtained as a processing result of the preprocessing unit 21, and extracts an image and position information for each character from the character group (step A3).

  Character extraction is performed, for example, by the following method. First, the projection in the horizontal direction is calculated for the image to be processed, and a place where the ratio between the value and the maximum value is equal to or less than a predetermined threshold is defined as a line delimiter. Then, the projection in the vertical direction is obtained within the separated line, and the place where the ratio of the value and the maximum value is equal to or less than a predetermined threshold is defined as the character separation, and the image inside the character rectangle thus obtained is represented by one character. Character image (first character image).

  Character extraction can also be performed as connected region extraction by labeling, and is not limited to the method described here, and any method may be used as long as it can acquire the image and position of the character.

  The character recognition unit 23 classifies the character image extracted by the character extraction unit 22 into one of predetermined classes by a method such as a composite similarity method or a subspace method, and a character code ( Character recognition is performed by outputting the first character (step A4). The character recognition means 23 basically outputs a character code corresponding to the input character image, but can also output a reject code for a character that is difficult to distinguish from the character image. Further, in methods such as the composite similarity method and the subspace method, the similarity value as a classification scale is used for class classification, but this similarity value may be output together.

  As shown in FIG. 3, the correction dictionary 25 (correction dictionary 25a) determines that the representative image (character image) of the pattern to be corrected is very similar to the character image to be processed. In this case, an entry made up of a set of data (for example, a character code) for specifying a character to be used as a correction result is registered. At least one entry is registered in the correction dictionary 25. Hereinafter, the image of the entry in the correction dictionary 25 is referred to as a reference image (second character image).

  The automatic correction means 24 determines whether or not the image input by the character recognition means 23 and the reference image in the correction dictionary 25 entry are very similar (step A5), and is determined to be very similar. If this happens (step A6, Yes), the output from the character recognition unit 23 is automatically corrected by replacing the output from the character recognition unit 23 with the correction result (character code (second character)) of the entry. (Step A7).

  Automatic correction may be performed when the reference image and the input character image are very similar, or when the character recognition unit 23 outputs the similarity calculated at the time of character recognition. It may be performed only when the value is lower than a predetermined threshold value. Alternatively, a list of characters having similar characters may be created in advance, and correction may be performed only when the list includes characters.

The determination of whether or not the character image is very similar is to determine a value representing the mismatch between the input character image and the reference image, and when the value exceeds a predetermined threshold, “not very similar” If it does not exceed, it is determined by determining that it is “similar”. As shown in FIG. 6, the value representing the degree of mismatch between the character images is obtained by translating and superimposing the reference image so that the centers of the circumscribed rectangles of the input character image and the reference image coincide with each other. The minimum rectangle including the circumscribed rectangle is calculated, the square error e calculated from the area A of the rectangle, the pixel value R of the reference image, and the pixel value I of the character image is calculated by the following equation (1), and the mismatch is determined. Used as a value to represent. However, the calculation of the sum in Expression (1) is performed for all pixels in the rectangle.

  Here, as a method of aligning the input character image and the reference image, a method using the center point of the circumscribed rectangle is shown, but in addition, a method of aligning the center of gravity of the black pixels of the character image and the reference image, A method of aligning the center of the lower side of the circumscribed rectangle as shown in FIG. 7, a method of aligning the center of the upper side of the circumscribed rectangle as shown in FIG. 8, and one of the corners of the circumscribed rectangle as shown in FIG. ) And a method of matching the reference image at a position where the square error e is minimized within a range where the circumscribed rectangle has a shared portion. Note that the alignment may be performed by any method as long as the value of the square error e can be expected to be small.

  Here, the square error e is used as a value representing the degree of mismatch between the aligned character image and the reference image. However, after the blurring process using Gaussian or the like, the square error e is obtained by the above-described method. Etc. are also conceivable. That is, any value may be used as long as it is a value calculated based on an image (pixel value) representing the degree of mismatch between the input image and the reference image.

  The manual correction means 26 presents the recognition result corrected by the automatic correction means 24 and the corresponding character image to the operator (step A8). The operator can correct the recognition result by inputting the character (third character) from the keyboard 15 or the like by looking at the presented character recognition result.

  FIG. 10 is a diagram illustrating an example of a recognition result and a character image presentation method (display screen). In the presentation method shown in FIG. 10, the character image extracted by the character extraction unit 22 and the recognition result corrected by the automatic correction unit 24 are displayed side by side. Also, in this example, an arrow-shaped cursor for selecting a character to be corrected by the operator is displayed, and the tip of the cursor is positioned on a specific character image or the recognition result area In addition, the character is selected.

  In the example of FIG. 10, the tip of the cursor is positioned in the area of the first character “” in the character image, and this character is selected.

  When the operator inputs a key from the keyboard 15 in a state where a specific character of the recognition result is selected (step A9, Yes), the character input by the key is input as a correction result corresponding to the character of the recognition result. . The manual correction means 26 corrects the corresponding character based on the input correction result (step A10), and outputs the corrected character as the recognition result 28.

  As shown in FIG. 11, the input of the character as the correction result is performed in an input field aligned with the recognition result. Alternatively, as shown in FIG. 12, the displayed recognition result may be corrected and directly edited.

  Next, the correction dictionary management means 27 executes correction dictionary management processing according to the flowchart shown in FIG. 5, and adds, changes, and deletes entries in the correction dictionary 25 based on corrections to the recognition result by the automatic correction means 24. (Step A11).

  The correction dictionary management means 27, when the character manually corrected by the manual correction means 26 is not registered as the correction character of the correction dictionary 25 (step B1, step B2, Yes), that is, the character corrected by the manual correction means 26 If an entry consisting of a pair of an image that is very similar to the image of the image and a correction result by the manual correction means 26 is not registered in the correction dictionary 25, an entry consisting of a pair of the character image and the correction result (character) is added. It adds to the correction dictionary 25 (step B3).

  The reference image stored in the entry to be added may be the character image itself extracted as a character recognition target by the character extraction unit 22, or image processing such as smoothing or resolution conversion on the image extracted by the character extraction unit 22. It may be an image subjected to.

  On the other hand, when the character manually corrected by the manual correction means 26 is registered as the correction character of the correction dictionary 25 (step B1, step B2, No), the correction dictionary management means 27 is corrected by the manual correction means 26. An entry consisting of a pair of a reference image that is similar to a character image but not determined to be very similar in the automatic correction means 24 and a recognition result in the manual correction means 26 is already in the correction dictionary 25. When registered, or when the entry is used by the automatic correction means 24, the entry (reference image) in the correction dictionary 25 is changed based on the character image that has been subjected to character recognition.

  The method for determining whether or not the character corrected by the manual correction means 26 and the reference image are similar is basically the same as the method for determining whether or not these images are very similar. The only difference from the method of determining whether or not they are very similar is that a value larger than that used in determining whether or not they are very similar is used as the threshold of the square error e.

  The reference image is changed by the same method as the alignment performed by determining whether or not the input character image and the reference image are very similar to each other, as shown in the following equation (2), For each pixel of the reference image, the average value R ′ of the old pixel value R of the reference image and the pixel value I of the character image is calculated, and the average value R ′ is used as the pixel value of the new reference image.

R ′ = (R + I) / 2 (2)
Further, as shown in FIG. 13, the correction dictionary 25 (correction dictionary 25b) includes a reference image change count counter N for each entry, and a new reference image is obtained by using the value of the change count calculated by the counter N. It is also possible to obtain the pixel value R ′ as a weighted average given by the following equation (3).

R ′ = (NR + I) / (N + 1) (3)
Note that the number of changes of the correction dictionary 25a is counted up when the correction dictionary management unit 27 executes a process of changing the entry.

  Further, if the data size of the correction dictionary 25 increases without limit, it is not preferable because of the limitation of the storage area. For this reason, the correction dictionary management means 27 can also delete the entry of the correction dictionary 25 and prevent an unlimited increase in the data size of the correction dictionary 25 (step B6).

  The entry deletion for such a purpose is performed by moving the added or changed entry to the head of the correction dictionary 25 every time an entry is added to the correction dictionary 25 or an entry in the correction dictionary 25 is changed. If the number of entries in the correction dictionary 25 exceeds a predetermined constant, this can be done by deleting the last entry. Alternatively, among a plurality of entries registered in the correction dictionary 25, entries that have not been referred to for the longest time in correction by the automatic correction means 24 may be deleted in order. Also, as shown in FIG. 13, when the number of changes is managed for each entry, the entry with the smallest number of changes can be deleted.

  Further, the manual correction means 26 displays the display form such as the display color and font of the recognition result of the character corrected by the automatic correction means 24 as shown in FIG. The operator can edit the entry of the correction dictionary 25 used for correction.

  As shown in FIG. 14, the operator corrects or deletes an entry by selecting a character that has been automatically corrected on the correction screen, as shown in FIG. 15, by key operation using the keyboard 15 or selection using a pointing device 16 such as a mouse. Do this by opening the menu and selecting the command displayed on it. The menu presented on the correction screen shown in FIG. 15 includes items “1. Delete entry”, “2. Change correction result”, and “3. Change reference image”.

  Here, when “1. Delete entry” is instructed by the operator, the correction dictionary management unit 27 deletes the entry corresponding to the selected character. Further, instead of actually deleting the entry from the correction dictionary 25, it is also possible to invalidate the entry by putting a deleted mark on the entry and ignoring the mark marked by the correction processing in the automatic correction means 24. Further, by removing the marked entry from the deletion target in the correction dictionary management means 27, the correction entry rejected by the user can be permanently invalidated.

  When “2. Modification Result Change” is instructed, the modification dictionary management unit 27 modifies the corresponding entry in the modification dictionary 25 according to the character code indicating the new modification result key-inputted from the keyboard 15. Change the result. In addition to direct key input, a list of similar characters can be presented and selected from them.

  When “3. Reference image change” is instructed, the correction dictionary management unit 27 can present a reference image editing screen or the like.

  By the way, the character recognition apparatus of the first embodiment may be configured such that a recognition dictionary 30 and a recognition dictionary management means 32 are added as shown in the functional block diagram of FIG.

  In the configuration shown in FIG. 16, the character recognition unit 23 executes character recognition processing using the recognition dictionary 30, and the recognition dictionary management unit 32 recognizes the recognition dictionary based on the correction results in the automatic correction unit 24 and the manual correction unit 26. 30 is corrected.

  The correction of the recognition dictionary 30 in the recognition dictionary management unit 32 may be performed by regularly performing re-learning using all of the learning pattern used for learning of the recognition dictionary 30 in the initial state and the corrected pattern. Then, additional learning for obtaining a learning result similar to that performed when relearning is performed may be performed each time the recognition result is corrected.

  Comparing the configuration in which the recognition dictionary management means 32 is added to the conventional method of FIG. 14 in which the recognition dictionary 30 is simply learned online, the automatic correction means 24, the correction dictionary 25, and the correction dictionary management within the dotted line in FIG. The means 27 is newly provided, and only a pattern very similar to the pattern corrected by these means can be automatically corrected.

  Thus, in the character recognition device of the first embodiment, when the character image to be processed is very similar to the reference image registered in the correction dictionary 25 with respect to the recognition result by the character recognition means 23. Can be automatically corrected by the character of the correction result of the corresponding entry. Accordingly, the recognition result for the character pattern (character image) that is very similar in image can be surely corrected by the entry in the correction dictionary 25. The entries in the correction dictionary 25 are added and changed based on the correction of the final recognition result according to the instruction from the operator by the manual correction means 26. Therefore, the correction to the recognition result by the operator is corrected. 25 can be reliably reflected. Further, the automatic correction means 24 automatically corrects a character that is very similar to a character that has been corrected in the past, thereby eliminating the need for the operator to repeatedly correct a character that is very similar to an image. Can be reduced.

(Second Embodiment)
The second embodiment of the present invention will be described below with reference to the drawings.
The character recognition device according to the second embodiment uses a character group (character string) as a modification to the character recognition result instead of a single character as in the first embodiment, and sets the entry in the correction dictionary 25 to a character string corresponding to the character group. Configure as.

  The character recognition device of the second embodiment is configured as shown in the functional block diagram of FIG. 2 as in the first embodiment. However, the preprocessing unit 21, the character extraction unit 22, the character recognition unit 23, and the manual correction unit 26 function in the same manner as in the first embodiment, and the automatic correction unit 24, the correction dictionary 25, and the correction dictionary management unit 27 are the first ones. Different from one embodiment.

  In the correction dictionary 25 (correction dictionary 25c) in the second embodiment, as shown in FIG. 17, a set of a representative image (character string image) of a pattern to be corrected and a character string of a correction result corresponding thereto. An entry consisting of is registered. The data indicating the character string is recorded as the correction result, which is different from the first embodiment in which a single character is stored as the correction result.

  Next, the operation of the character recognition device in the second embodiment will be described with reference to the flowcharts shown in FIGS.

  The processing flow of the second embodiment shown in FIG. 18 is the same as that of the first embodiment shown in FIG. 4, and steps A1 to A11 correspond to C1 to C11, respectively. Description of the same processing is omitted, and processing different from that of the first embodiment is described below.

  Whether the reference image in the entry of the correction dictionary 25 is very similar to the image of the character group subjected to character recognition processing by the character recognition unit 23 having the same number of characters as the corrected character string length in the entry. (Step C6), and if it is determined that they are very similar (step C6, Yes), character recognition is performed by replacing the output from the character recognition means 23 with the correction result (character string code) of the entry. The output from the means 23 is automatically corrected (step C7).

  The automatic correction by the automatic correction unit 24 may be performed when the reference image and the character group image to be recognized are very similar, or the character recognition unit 23 outputs the similarity. May be performed only when the set of similarity in the character group satisfies a predetermined condition, such as when the average value of the values is lower than a predetermined threshold. Alternatively, a list of characters having similar characters may be created in advance, and correction may be performed only when the characters included in the list exist in the character group.

  As a method for determining whether or not the character group image and the reference image are very similar, the same method as that used in the automatic correction means 24 of the first embodiment is used.

  Next, the correction dictionary management unit 27 executes a correction dictionary management process according to the flowchart shown in FIG. 19, and adds, changes, and deletes entries in the correction dictionary 25 based on the correction to the recognition result by the automatic correction unit 24. (Step C11).

  First, the correction dictionary management unit 27 lists the characters corrected by the manual correction unit 26, the character group including the character group and the preceding and following character groups, the number of characters being within a predetermined number M of characters. (Step D1). For example, when M = 2 and the character “t” (represented by a white character in the figure) is corrected in the character string of FIG. 20, the three character groups A, B, and C of FIG. 21 are listed. The

  Then, the correction dictionary management unit 27 determines, for each of the enumerated character groups, whether or not there is an entry including a set of an image that is very similar to the image of the character group and the correction result (step D2).

  When no entry satisfying the above condition is found in the correction dictionary 25 (step D3, Yes), the correction dictionary management unit 27 includes a character group image including characters corrected by the manual correction unit 26, and a manual correction unit. An entry made up of a pair of character strings (correction results) including the characters corrected by 26 is added to the correction dictionary 25.

  Further, the entry in the correction timeline is changed when an entry satisfying the above condition is found in the correction dictionary 25 or when the entry is used by the automatic correction means 24. The change of the reference image and the deletion of the modified dictionary 25 entry are performed by the same method as in the first embodiment (step D3, step D5).

  Further, in the case where the recognition dictionary 30 is used in the character recognition means 23, a recognition dictionary management means 32 for correcting the recognition dictionary 30 based on the correction results in the automatic correction means 24 and the manual correction means 26 is added. The points that may be configured as described above are the same as in the first embodiment.

  Thus, in the second embodiment, the correction unit in the automatic correction means 24 is not a single character but a character group (character string), and the entry in the correction dictionary 25 is managed as a set of character strings corresponding to the character group. be able to. By allowing a character string made up of a plurality of characters as a unit for automatic correction, the result of automatic correction can be changed depending on the pattern of adjacent characters even for characters that are very similar in terms of characters.

(Third embodiment)
The third embodiment of the present invention will be described below with reference to the drawings.
FIG. 22 is a block diagram showing the configuration of the character recognition device in the third embodiment.
In the character recognition device of the third embodiment, a word correction unit 40 is provided between the character recognition unit 23 and the automatic correction unit 24. The word correcting means 40 searches the recognition result for a character string similar to the word in the word dictionary 42, and replaces the mismatched portion with the word in the character string similar to the word with the corresponding portion of the word. Correct it. By providing the word correction means 40, the correction dictionary management means 27 adds and changes entries to the correction dictionary 25 not only for characters corrected by the manual correction means 26 but also for characters corrected by the word correction means 40. Also do about.

  In FIG. 22, the preprocessing means 21, the character extraction means 22, the character recognition means 23, and the manual correction means 26 function in the same way as in the second embodiment, and words expected to be included in the recognition target document are shown. The second embodiment is different from the second embodiment in that a registered word dictionary 42 and word correction means 40 for correcting the recognition result of the character recognition means 23 using the word dictionary 42 are added. In the third embodiment, the correction by the automatic correction unit 24 is performed not on the processing result of the character recognition unit 23 but on the processing result of the word correction unit 40, and the correction dictionary management unit 27 also includes information on the correction in the word correction unit 40. Use.

  The word dictionary 42 stores words that are expected to be included in the document to be recognized. For example, when a Japanese sentence is a recognition target, a Japanese language dictionary is used, and when an English sentence is a recognition target, An English word dictionary 42 is stored.

  Next, the operation of the character recognition device in the third embodiment will be described with reference to the flowchart shown in FIG.

The process flow of the third embodiment shown in FIG. 23 is substantially the same as that of the second embodiment shown in FIG. 18, and steps C1 to C4 and C5 to C11 correspond to E1 to E4 and E6 to E12, respectively. ing. Description of the same processing is omitted, and processing different from that of the second embodiment is described below.

The word correction means 40 uses a dynamic programming method or the like to make a character string similar to the word character string in the word dictionary 42 out of the character string as a result of processing by the character recognition means 23, the number of characters not equal to the word character string. Is searched for a character string that is equal to or less than a predetermined constant for each word length, and the character string, the corresponding word, and its position are specified . When a character string satisfying the above conditions is found and there is a mismatched character in the character string and the word, the recognition result is corrected by replacing the character in the mismatched portion with the character in the corresponding portion of the word (step E5).

  The method of adding and changing entries to the correction dictionary 25 by the correction dictionary management means 27 is basically executed in the same manner as in the second embodiment (FIG. 19).

  However, the correction dictionary management means 27 of the third embodiment performs addition and change of entries not only for characters corrected by the manual correction means 26 but also for characters corrected by the word correction means 40.

  That is, in steps D1 and D4 of the flowchart shown in FIG. 19, not only the manually corrected characters but also the word corrected character strings (character groups) are processed in the same manner.

  If the recognition dictionary 30 is used in the character recognition means 23, a recognition dictionary management means 32 for correcting the recognition dictionary 30 based on the correction results in the automatic correction means 24 and the manual correction means 26 may be added. The good point is the same as in the first embodiment.

In this way, in the third embodiment, not only the words in the word dictionary 42 but also the partial character strings of the words in the word dictionary are corrected by the entry added by the correction by the word correcting means 40. The number of corrections by the operator can be reduced and the work load can be reduced as compared with the case where correction is simply performed by the word correction means 40.

  In addition, this invention is not limited to embodiment mentioned above, In the implementation stage, it can change variously in the range which does not deviate from the summary. Further, the functions executed in the above-described embodiments may be combined as appropriate as possible. The above-described embodiments include various stages of the invention, and various inventions can be extracted by appropriately combining a plurality of disclosed constituent elements. For example, even if several constituent requirements are deleted from all the constituent requirements shown in the embodiment, an effect can be obtained, so that a configuration from which the constituent requirements are deleted can be extracted as an invention.

  The processing described in each of the embodiments described above is a character recognition program that can be executed by a computer, such as a magnetic disk (flexible disk, hard disk, etc.), an optical disk (CD-ROM, DVD, etc.), and a semiconductor memory. It can be written on a recording medium and provided to various apparatuses. It is also possible to transmit to a variety of devices by transmitting via a communication medium. The computer that implements the character recognition device reads the character recognition program recorded on the recording medium or receives the character recognition program via the communication medium, and the operation is controlled by the program to execute the above-described processing. To do.

The block diagram which shows the structure of the system which implement | achieves the character recognition apparatus in embodiment of this invention. The block diagram which shows the function structure of the character recognition apparatus in 1st Embodiment implement | achieved by the system shown in FIG. The figure which shows an example of the entry registered into the correction dictionary 25 in 1st Embodiment. The flowchart for demonstrating operation | movement (character recognition process) of the character recognition apparatus in 1st Embodiment. The flowchart for demonstrating operation | movement (correction dictionary management process) of the character recognition apparatus in 1st Embodiment. The figure for demonstrating the superimposition of the image for calculating | requiring the value showing the mismatch of the character image and reference image in 1st Embodiment. The figure for demonstrating the superimposition of the image for calculating | requiring the value showing the mismatch of the character image and reference image in 1st Embodiment. The figure for demonstrating the superimposition of the image for calculating | requiring the value showing the mismatch of the character image and reference image in 1st Embodiment. The figure for demonstrating the superimposition of the image for calculating | requiring the value showing the mismatch of the character image and reference image in 1st Embodiment. The figure which shows an example of the presentation method (display screen) of the recognition result and character image in 1st Embodiment. The figure which shows an example of the presentation method (display screen) of the recognition result and character image in 1st Embodiment. The figure which shows an example of the presentation method (display screen) of the recognition result and character image in 1st Embodiment. The figure which shows an example of the entry of the correction dictionary 25 provided with the counter of the reference image change frequency in 1st Embodiment. The figure which shows an example of the correction screen for editing the entry of the correction dictionary 25 in 1st Embodiment. The figure which shows an example of the correction screen for editing the entry of the correction dictionary 25 in 1st Embodiment. The block diagram which shows the function structure which added the recognition dictionary 30 and the recognition dictionary management means 32 in 1st Embodiment. The figure which shows an example of the entry of the correction dictionary 25 in 2nd Embodiment. The flowchart for demonstrating operation | movement (character recognition process) of the character recognition apparatus in 2nd Embodiment. The flowchart for demonstrating operation | movement (correction dictionary management process) of the character recognition apparatus in 2nd Embodiment. The figure which shows an example of the corrected character string in 2nd Embodiment. The figure which shows an example of the character group enumerated with respect to the correction character shown in FIG. The block diagram which shows the function structure of the character recognition apparatus in 3rd Embodiment. The flowchart for demonstrating operation | movement (character recognition process) of the character recognition apparatus in 3rd Embodiment.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 10 ... CPU, 12 ... Memory, 12a ... Character recognition program, 12b ... Original image data, 12c ... Correction dictionary data, 12d ... Character recognition result data, 14 ... Display, 15 ... Keyboard, 16 ... Pointing device, 17 ... Scanner, DESCRIPTION OF SYMBOLS 18 ... Memory | storage device, 19 ... Printer, 20 ... Original image, 21 ... Pre-processing means, 22 ... Character extraction means, 23 ... Character recognition means, 24 ... Automatic correction means, 25 ... Correction dictionary, 26 ... Manual correction means, 27 ... correction dictionary management means, 28 ... recognition result, 30 ... recognition dictionary, 32 ... recognition dictionary management means, 40 ... word correction means, 42 ... word dictionary.

Claims (10)

Character recognition means for recognizing a first character image and outputting a first character as a recognition result;
A modified dictionary in which an entry including a pair of data specifying a second character image and a second character corresponding to the second character image is recorded;
The first character image and the second character image are overlapped, and it is determined whether or not the first character image and the second character image are very similar by comparing pixel values of respective pixels. Discriminating means to perform,
When it is determined by the determining means that the first character image and the second character image are very similar, the first character as the recognition result output by the character recognizing means is the first character. a first character modification means for modifying the second letter indicating the data specifying the second character image and paired the character,
Presenting means for presenting the second character corrected by the first character correcting means as an object of manual correction;
Input means for inputting a third character for correcting the second character presented by the presenting means by a manual correction operation;
A second character modifying means for modifying the second character presented by the presenting means, the third characters input by said input means,
When the data for specifying the third character corrected by the second character correcting means is not recorded in the correction dictionary, the first character image and the data for specifying the third character are And a correction dictionary managing means for adding the entry to the correction dictionary . The discrimination means includes
A value representing a degree of mismatch is calculated based on pixel values of the first character image and the second character image, and the first character image is determined when the value does not exceed a predetermined threshold value. The character recognition apparatus according to claim 1, wherein it is determined that the second character image is very similar to the second character image. The correction dictionary management means includes
When data specifying the third character corrected by the second character correcting means is recorded in the correction dictionary, the second image that is a pair of data specifying the third character is The character recognition device according to claim 1, wherein the character recognition device is changed based on the first image. The correction dictionary management means includes
The first image and the second image are overlapped to calculate an average value of pixel values for each pixel, and the average value is used as a pixel value after the change of the second image. The character recognition device according to claim 4. In the correction dictionary, the number of changes by the correction dictionary management means for the entry is recorded,
The correction dictionary management means includes
The first image and the second image are overlapped to calculate an average value of pixel values weighted by the number of changes for each pixel, and the average value is used as a pixel value after the change of the second image The character recognition device according to claim 4, wherein: Character recognition means for recognizing the first character image and outputting a character as a recognition result;
A modified dictionary in which an entry consisting of a pair of data specifying a character string image and a character string is recorded;
Determining whether or not the first character image and the character string image are very similar by superimposing a plurality of the first character images and the character string image and comparing pixel values of the respective pixels Means,
A first character composed of a plurality of characters output as a recognition result by the character recognition unit when the determination unit determines that the plurality of first character images and the character string image are very similar. First character string correcting means for correcting a column to a second character string indicated by data specifying the character string paired with the character string image;
Input means for inputting a third character for correcting the second character string corrected by the first character string correcting means by a manual correction operation;
A second character string correcting means for correcting the third string containing the first of said said second string has been modified by the character string correcting means is input by said input means third letter,
An entry consisting of a pair of an image very similar to the image of the character group including the third character in the third character string and data specifying the character string corresponding to the character group is registered in the correction dictionary. A correction dictionary management means for adding, to the correction dictionary, an entry consisting of a character string image of the character group including the third character and a data pair specifying the character string corresponding to the character group when there is not A character recognition device. Character recognition means for recognizing the first character image and outputting a character as a recognition result;
A modified dictionary in which an entry consisting of a pair of data specifying a character string image and a character string is recorded;
A word dictionary in which words consisting of multiple characters are registered,
Word correction means for correcting a character string composed of a plurality of characters output by the character recognition means based on words registered in the word dictionary;
A plurality of the first character images corresponding to the character string to be corrected by the word correcting means are superimposed on the character string image, and a plurality of the first character images are compared by comparing pixel values of respective pixels. Determining means for determining whether or not the character image and the character string image are very similar;
When it is determined that the plurality of first character images and the character string image are very similar by the determining means, the first character string corrected by the word correcting means is compared with the character string image. First character string correcting means for correcting the second character string indicated by the data for specifying the character string ,
Input means for inputting a third character for correcting the second character string corrected by the first character string correcting means by a manual correction operation;
A second character string correcting means for correcting the third string containing the first of said said second string has been modified by the character string correcting means is input by said input means third letter,
An entry consisting of a pair of an image very similar to the image of the character group including the third character in the third character string and data specifying the character string corresponding to the character group is registered in the correction dictionary. A correction dictionary management means for adding, to the correction dictionary, an entry consisting of a character string image of the character group including the third character and a data pair specifying the character string corresponding to the character group when there is not A character recognition device. Computer
Character recognition means for recognizing a first character image and outputting a first character as a recognition result;
A modified dictionary in which an entry including a pair of data specifying a second character image and a second character corresponding to the second character image is recorded;
The first character image and the second character image are overlapped, and it is determined whether or not the first character image and the second character image are very similar by comparing pixel values of respective pixels. Discriminating means to perform,
When it is determined by the determining means that the first character image and the second character image are very similar, the first character as the recognition result output by the character recognizing means is the first character. a first character modification means for modifying the second letter indicating the data specifying the second character image and paired the character,
Presenting means for presenting the second character corrected by the first character correcting means as an object of manual correction;
Input means for inputting a third character for correcting the second character presented by the presenting means by a manual correction operation;
Second character correcting means for correcting the second character presented by the presenting means to the third character inputted by the input means ;
When the data for specifying the third character corrected by the second character correcting means is not recorded in the correction dictionary, the first character image and the data for specifying the third character are The character recognition program for functioning as a correction dictionary management means to add the entry to the correction dictionary . Computer
Character recognition means for recognizing the first character image and outputting a character as a recognition result;
A modified dictionary in which an entry consisting of a pair of data specifying a character string image and a character string is recorded;
A plurality of the first character images and the character string image are overlapped, and a determination is made as to whether or not the plurality of first character images and the character string image are very similar by comparing pixel values of respective pixels. Discriminating means to perform,
A first character composed of a plurality of characters output as a recognition result by the character recognition unit when the determination unit determines that the plurality of first character images and the character string image are very similar. First character string correcting means for correcting a column to a second character string indicated by data specifying the character string paired with the character string image;
Input means for inputting a third character for correcting the second character string corrected by the first character string correcting means by a manual correction operation;
A second character string correcting means for correcting the third string containing the first of said said second string has been modified by the character string correcting means is input by said input means third letter,
An entry consisting of a pair of an image very similar to the image of the character group including the third character in the third character string and data specifying the character string corresponding to the character group is registered in the correction dictionary. In order to function as a correction dictionary management means for adding an entry consisting of a character string image of a character group including the third character and a data pair specifying a character string corresponding to the character group to the correction dictionary when there is not Character recognition program. Computer
Character recognition means for recognizing the first character image and outputting a character as a recognition result;
A modified dictionary in which an entry consisting of a pair of data specifying a character string image and a character string is recorded;
A word dictionary in which words consisting of multiple characters are registered,
Word correction means for correcting a character string composed of a plurality of characters output by the character recognition means based on words registered in the word dictionary;
A plurality of the first character images corresponding to the character string to be corrected by the word correcting means are superimposed on the character string image, and a plurality of the first character images are compared by comparing pixel values of respective pixels. Determining means for determining whether or not the character image and the character string image are very similar;
When it is determined that the plurality of first character images and the character string image are very similar by the determining means, the first character string corrected by the word correcting means is compared with the character string image. First character string correcting means for correcting the second character string indicated by the data for specifying the character string ,
Input means for inputting a third character for correcting the second character string corrected by the first character string correcting means by a manual correction operation;
A second string modifying means for modifying the second character string that has been modified to a third string including the third characters input by said input means by the first character string correcting means,
An entry consisting of a pair of an image very similar to the image of the character group including the third character in the third character string and data specifying the character string corresponding to the character group is registered in the correction dictionary. In order to function as a correction dictionary management means for adding an entry consisting of a character string image of a character group including the third character and a data pair specifying a character string corresponding to the character group to the correction dictionary when there is not Character recognition program.

Images