JP2001147988A - Method and device for recognizing character - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a method and a device for recognizing character, with which the rate of recognition is improved in the case of reading a document or the like with ruled lines or field lines printed in one non-dropout color with an image reader. SOLUTION: A character is segmented from a read image, from which ruled lines or the like are removed (S1 and S2), ruled line contact information is stored (S3), recognized candidate information is provided by recognizing processing (S4 and S5), the group of corrected images Ci2 is provided by repeating operation for correcting that segmented image close to a form, which is predicted from a code Co1 and a distance Di1 of candidate characters provided by recognizing processing and registered form information Fin, just for the number of predicted forms (S6), the corrected image of high reliability is provided by discriminating the reliability of these corrected images (S7) and further, it is discriminated whether or not the recognized result is to be outputted (S8).

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character recognition technology, and more particularly to a character recognition technology which reads a character written on a form on which a ruled line and a frame line are printed by one color printing of a non-dropout color with an image reader. The present invention relates to a technology for recognizing characters from which ruled lines or frame lines have been removed.

[0002]

2. Description of the Related Art An image reader such as an OCR or a scanner reads a character on a form or a document, converts the character into an electric signal, and outputs a character image. However, a form usually has a ruled line or a frame line printed thereon. Are printed or written along the ruled lines, so that the ruled lines and frame lines are printed in a dropout color so as not to interfere with reading. Therefore, even if a form on which characters are printed or written is read by an image reader, the ruled lines or the frame lines are not read, so that the characters without those ruled lines or the frame line images, that is, only the characters can be read.

However, conventionally, in order to perform character recognition processing by reading such a form with an image reader, a non-dropout color guide mark indicating the position of a ruled line or a frame line is printed in order to determine the character entry position. Had to be kept.

However, in the method of printing the guide mark of the non-dropout color as described above, it is necessary to print the form in two colors of the ruled line or frame of the dropout color and the guideline of the non-dropout color. There is a problem that printing cost is high and running cost is high.

[0005]

In order to avoid an increase in running cost due to the use of the above-described two-color printing form, a form printed with non-dropout color one-color printing may be used. If a ruled line or frame printed in a non-dropout color touches or overlaps with a written character, the ruled line or frame cannot be distinguished from the character when read by an image reader, resulting in erroneous recognition or inability to read. There was a problem that there was a case.

Therefore, in order to prevent erroneous recognition or inability to read when a character contacts a ruled line or a frame line, the ruled line or the frame line may be forcibly removed after being read by an image reader. Since the recognition rate of the character image after the removal is reduced only by removing the ruled line or the frame line (since a part of the character in contact with the ruled line or the frame line is removed), conventionally, the remaining character The image correction is performed by estimating the removed portion from the directions of the strokes before and after the image portion and the distance thereof.

However, there is a problem that correct image correction cannot be performed when the direction of the stroke is different from the true character image in the above-described image correction method. For example, if the lower part of the character "7" is in contact with the ruled line as in the example of FIG. 3, the information is lost together with the ruled line, and the stroke direction is diagonally lower left as in the example of FIG. Since the distance is long and the distance is long, if the image is corrected by the conventional method, an image in which the stroke portion is extended diagonally to the left as shown in FIG. 10C ("7" if character recognition is performed) is obtained, and a correct character image ( FIG.
Image correction different from (b)) is performed.

SUMMARY OF THE INVENTION The present invention has been made to solve the above-mentioned problem, and is intended to read a form or the like on which a ruled line or a frame (hereinafter referred to as a ruled line) is printed by non-dropout color printing with an image reader. It is an object of the present invention to provide a character recognition method and a character recognition device which realize an improvement in the recognition rate of a character.

[0009]

In order to solve the above-mentioned problems, a character recognition method according to a first aspect of the present invention is to remove a ruled line or the like from a read image of a read original and cut out a character image one by one to extract a character image. This is a character recognition method for performing recognition, comprising obtaining contact information between a character image cut out at the time of cutting out the character image and a ruled line, performing recognition processing of the cut out character, and obtaining the recognition processing. Based on the recognized recognition result and the contact information, a corrected image group is obtained by correcting the cut-out character image so as to approximate a shape predicted from the pre-registered predicted shape information, and a recognition candidate is selected from the corrected image group. It is characterized in that an image is selected and whether or not to output a recognition character corresponding to the recognition candidate image is evaluated.

A character recognition apparatus according to a second aspect of the present invention is a character recognition method for removing characters such as ruled lines from a read image of a read original and cutting out character images one by one to perform character recognition. The contact information between the character image cut out at the time of cutting out and the ruled line, etc. is obtained, and recognition processing of the cut out character is performed.
Based on the recognition result obtained in the recognition processing and the contact information, a corrected image group is obtained by correcting the cut-out character image so as to approximate a shape predicted from the pre-registered predicted shape information. Check whether the corrected image in the image satisfies a predetermined condition, and extract a recognition candidate image from the corrected image satisfying the predetermined condition,
It is characterized in that whether or not to output a recognition character corresponding to the recognition candidate image is evaluated.

According to a third aspect of the present invention, in the character recognition method according to the second aspect of the present invention, when all of the corrected images in the corrected image group do not satisfy a predetermined condition, the character image recognition result further includes On the basis of the information, a cut-out character image is obtained in a predetermined correction priority order to obtain a corrected image group corrected so as to approach a shape predicted from pre-registered predicted shape information. It is characterized in that the operation of checking whether or not the correction image in the image satisfies a predetermined condition and extracting a recognition candidate image from the correction image satisfying the predetermined condition is repeated.

According to a fourth aspect of the present invention, there is provided a character recognition device for removing ruled lines and the like from a read image of a read original, and one character at a time from the character image from which the ruled lines and the like have been removed by the ruled line removing means. A cutout means for cutting out a character image, a ruled line contact information obtaining means for obtaining contact information of a portion where a ruled line or the like has contacted from a character image from which a ruled line or the like has been removed by a ruled line removing means, and a character image cut out by the cutout means Based on the recognition result obtained by the recognition process and the contact information obtained by the recognition process, and correcting the cut-out character image to approximate the shape predicted from the pre-registered predicted shape information. Image correction means for obtaining a group of corrected images, and a correction image obtained by the image correction means. Correction image selection means for selecting a recognition candidate image from a group of images, and recognition character output evaluation means for evaluating whether to output a recognition character corresponding to the recognition candidate image obtained by the correction image selection means. , Is provided.

According to a fifth aspect of the present invention, there is provided a character recognition device for removing ruled lines and the like from a read image of a read original, and one character at a time from the character image from which the ruled lines and the like have been removed by the ruled line removing means. A cutout means for cutting out a character image, a ruled line contact information obtaining means for obtaining contact information of a portion where a ruled line or the like has contacted from a character image from which a ruled line or the like has been removed by a ruled line removing means, and a character image cut out by the cutout means Based on the recognition result obtained by the recognition process and the contact information obtained by the recognition process, and correcting the cut-out character image to approximate the shape predicted from the pre-registered predicted shape information. Image correcting means for obtaining a corrected image group and characters after correction by the character image correcting means Determining whether the image satisfies a predetermined condition, and extracting a recognition candidate image from the correction images satisfying the predetermined condition; and a recognition candidate image obtained by the correction image judgment unit. And a recognition character output evaluating means for evaluating whether or not to output a corresponding recognition character.

According to a sixth aspect of the present invention, in the character recognition device according to the fifth aspect of the present invention, the image correcting means determines whether or not all the corrected images satisfy the predetermined condition as a result of the determination by the corrected image determining means. Means for correcting the cut-out character image based on the recognition result of the image and the contact information so that the cut-out character image approaches a shape predicted from pre-registered predicted shape information according to a predetermined correction priority. , Characterized in that.

[0015]

FIG. 1 is a block diagram showing the configuration of an embodiment of a character recognition apparatus according to the present invention.
Reference numeral 00 includes a ruled line removing unit 10, a character cutout unit 20, a ruled line contact information storage unit 30, and a character recognition block 40. Although not shown, the character recognition device 100
Has a control unit composed of a CPU and its peripheral circuits, which controls the operation of each of the above-described components and controls the operation of the entire character recognition apparatus.

The ruled line removing section 10 detects a ruled line or the like (ruled line or frame line) from the read image Im1 (FIG. 3) read from a scanner or the like and acquires a character image Im2 (FIG. 4) from which the image of the ruled line or the like has been removed. Then, rule line contact information (for example, the contact direction of the character, the number of contact points, the position (coordinates) of the contact portion, the thickness of the rule line, etc.) If1 of the image touching the rule line or the like is acquired.

The character cutout unit 20 cuts out character images one by one from the image Im2 from which the ruled lines and the like have been removed by the ruled line removing unit 10, and outputs the cutout character images C.
i1 is obtained, and the ruled line contact information of the cut-out character image is converted into ruled line contact information If2 for each character.

The ruled line contact information storage unit 30 comprises a temporary storage memory such as a RAM, and stores the ruled line contact information If2 obtained by the character cutout unit 20.

The character recognition block 40 includes a feature extraction section 41, a dictionary section 42, an identification section 43, and a predicted shape information section 4.
4. Image correction processing unit 45, correction image selection unit 46
And a rejection determination unit 47 for performing character extraction after performing feature extraction, image correction processing, and the like.

That is, in the character recognition block 40, the feature extracting unit 41 calculates a feature amount Fd1 from the character image Ci1 cut out by the character cutting unit 20.

The dictionary unit 42 stores standard feature values Fdd, such as numbers and alphabets, and character codes Co.
A plurality of templates are stored in advance in a storage memory such as a ROM for each character type.

The discriminating unit 43 calculates the distance between the feature value Fd1 calculated by the feature extracting unit 41 and the feature value Fdd of each template of the dictionary unit 42. Character code Co from to upper specified value
1. Acquire recognition candidate information such as the distance Di1.

The predicted shape information unit 44 stores information Fin of all possible shapes depending on the state of contact with a ruled line or the like in the RO.
M and stored in a storage memory such as M.

The image correction processing section 45 stores the code Co1 and the distance Di1 of the candidate character obtained by the identification section 43 corresponding to the character image Ci1 cut out by the character cutout section 20, and the ruled line contact information. Part 3
On the basis of the ruled line contact information fi2 stored in 0, the number of operations to correct the shape to be close to the shape predicted from the shape information Fin stored in the predicted shape information unit 44 is the number of all predicted shapes. By repeating this, a group of corrected images Ci2 is obtained.

Further, the correction image selection section 46 controls all the correction images Ci corrected by the image correction processing section 45.
A reliable corrected character image Ci3 is selected from the two groups.

The rejection determination unit 47 determines the reliability of the corrected character image Ci3 selected by the correction image selection unit 46 as a character. If the reliability is higher than a low value, the character code is output. Otherwise, it outputs a reject code (rejection code).

FIG. 2 is a flowchart showing an example of a character recognition operation performed by the character recognition apparatus 100 shown in FIG. 1. The control of the operation sequence of each step is performed by the control unit.
FIG. 3 is a diagram showing an example of a read character image. 4 is a diagram showing an example of a character image after removing ruled lines and the like, and FIG. 5 is a diagram showing a corrected image and a recognition result of the character “9” after removing ruled lines and the like in FIG. 4 as an example. It is.

Step S1: (Removal of Ruled Lines, etc.) In FIG. 2, the ruled line removing unit 10 is a non-dropout color form or document (character, "2" in the example of FIG. 3) taken into a temporary storage memory such as a DRAM. , "9"), a character image Im2 (FIG. 4) from which the ruled lines and the like (ruled lines denoted by reference numerals 31 and 32 and frame line denoted by reference numeral 33 in the example of FIG. 3) of the read image Im1 are removed. Get
It is stored in a temporary storage memory such as a DRAM.

Step S2: (Cutout of Character) Next, the character cutout unit 20 cuts out character images one by one from the image Im2 from which the ruled lines and the like have been removed in step S1, and cuts out the character image Ci1 (see FIG. 4). Then, each cut-out character image represented by reference numerals 41 and 42 is obtained.

Step S3: (Acquisition and storage of ruled line contact information) The character cutout unit 20 converts the ruled line contact information of the cut-out character image cut out in step S2 into ruled line contact information If2 for each character (for example, character contact information). The direction, the number of contact points, the position (coordinates) of the contact part, the thickness of the ruled line of the contact part, etc.) are converted (in the example of FIG. 4, two pieces of ruled line contact information of the cut-out character images 41 and 42 are obtained). The information is stored in the information storage unit 30 (when there is a character that is not in contact with a ruled line or the like, ruled line contact information (for example, the number of contact points = 0) meaning "no ruled line contact" is stored). For example,
If the character image represented by reference numeral 42 in the two-character cutout image shown in FIG. 4 is taken as an example, the character “9” in FIG.
Is in contact with the right ruled line 32 at three portions 51, 52, and 53, and if the ruled line is removed in step S1, a cutout image as indicated by reference numeral 42 is obtained (that is, FIG. One character image is cut out with one place open as shown). Also, in this example,
The character cutout unit 20 determines the position of the ruled line 32 in contact with the character image 42 as the ruled line contact information of the character image (that is, the positions 51, 5, 5
2 and the position 53 where the character “9” contacts below the ruled line 32
Is stored in the ruled line contact information storage unit 30. In the embodiment, the positions 51, 52, and 53 are represented by coordinate values (X, Y), but the present invention is not limited to this.

Step S4: (Characteristic Extraction of Cut-out Character Image) The feature extracting unit 41 extracts the 1
Feature extraction is performed from one character image Ci1 (for example, the cut-out character image in FIG. 5) to calculate a feature amount Fd1.

Step S5: (Acquisition of Recognition Candidate Information of Cut-out Character Image) Next, the identification unit 43 stores the feature amount Fd1 of the character image Ci1 calculated in step S4 and the templates of the dictionary unit 42. The distance calculation is performed with respect to the standard feature value Fdd, and the recognition candidate character codes Co1 and the distance calculation result Di1 from the smallest distance (the order of the closest feature) to the top three are acquired as recognition candidate information (for example, As a result of performing the identification processing of step S5 on the cut-out character image in FIG. 5, the first and second recognition candidate character codes of the first and second recognition candidate character codes Co1 up to the third recognition position are “3”. The third recognition candidate character code is a character code, and a recognition character code such as “9” is obtained. Fruit is obtained). In the embodiment, the number of the recognized character codes Co1 to be obtained is three (first to third), but the number of the recognized character codes Co1 is not limited to this.

Step S6: (Acquisition of Corrected Image Group) In the image correction processing unit 45, the shape information Fin stored in the predicted shape information unit 44 and the conditions acquired by the operation up to this point (cut-out character image (for example, character Image 42) ruled line contact information If2, recognition candidate character candidate Co1
And the predicted shape information section 44 based on the distance calculation result Di1))
To obtain the shape information that satisfies the conditions obtained by the operations up to here. Next, for the target cut-out character image, based on the code Co1 and distance Di1 of the candidate character obtained by the corresponding identification unit 43 and the ruled line contact information fi2 stored in the ruled line contact information storage unit 30, The operation of correcting the shape to be closer to the shape predicted from the shape information Fin stored in the predicted shape information unit 44 is repeated by the number of all predicted shapes to obtain a group of corrected images Ci2 (for example, FIG. In the case of a cut-out character image, a correction image when there is a possibility of the character “3” from the predicted shape information unit 44 (FIG. 6A) and a correction image when there is a possibility of the character “8” (FIG. 6 (b)) and a corrected image (FIG. 6 (c)) when there is a possibility of the character "9".

Step S7: (Correction Image Selection Processing) The correction image selection section 46 selects the correction image Ci obtained by the image correction processing section 45 in step S6.
Of the two groups (the three types of corrected images Ci2 described above in the case of the character image in FIG. 5), the corrected image having the shape with the highest reliability is selected as the corrected image Ci3 by a predetermined selection method. In this case,
In the embodiment, a correction image is selected by performing a process of identifying each correction image Ci2. That is, the feature amount of each correction image Ci2 is calculated, and the standard feature amount Fdd stored in each template of the dictionary unit 42 is calculated.
Is calculated, and based on the calculation result, a corrected image Ci3 having the smallest distance and the farthest distance from another category among the corrected images Ci2 is selected (in the example of FIG. 5, the above-described three types are selected). Among the correction images of FIG. 6C, the correction image of FIG. 6C which is to be corrected by approaching the shape of “9” having the smallest distance and being farthest from other categories (that is, having the highest reliability) is obtained. Selected). Note that the selection method is not limited to this method.

Step S8: (Judgment of Reliability as Character) Next, the rejection judging section 47 selects the corrected image Ci3 selected by the corrected image selecting section 46 in step S7.
Is calculated with respect to the standard feature value Fdd stored in each template of the dictionary unit 42, and the distance r
And a predetermined threshold δ, and r−δ <0 (that is, distance <
In the case of (threshold value), it is evaluated that there is absolute reliability, and the process proceeds to S9; otherwise, the process proceeds to S10.

Step S9: (Output of Character Recognition Code) If the distance from the feature value Fdd is smaller than the threshold value, the control unit converts the character code obtained from the dictionary unit 42 in step S7 into a recognition character This is output as a code, and the recognition process for one character is completed.

Step S10: (Output of Reject Code) When the distance from the feature value Fdd is smaller than the threshold value, the control unit outputs a reject code and ends the recognition processing for one character.

With the above configuration, the processing for removing ruled lines and the like (S
According to 1), even if true image information is lost as shown in FIG. 5, a corrected image as shown in FIG. 6C can be obtained and a highly reliable recognition result can be output.

FIG. 7 is a block diagram showing the configuration of another embodiment of the character recognition device according to the present invention. The reliability of the acquired corrected image is determined by a predetermined priority (priority based on the order in which correction is performed). 2), when the reliability of the corrected image is low, the operation of checking the reliability of the corrected image of the next rank is repeated to obtain a highly reliable corrected image as a character image.

In FIG. 7, the character recognition device 100 'includes a ruled line removing unit 10, a character cutout unit 20, and a ruled line contact information storage unit 3.
0 and a character recognition block 40 '. Although not shown like the character recognition device 100 of FIG. 1, the character recognition device 100 'is composed of a CPU and its peripheral circuits, and controls the operation of each of these components and controls the operation of the entire character recognition device. It has.

Here, the configurations, functions, and operations of the ruled line removing unit 10, the character cutout unit 20, and the ruled line contact information storage unit 30 are the same as those of the character recognition device 100 of FIG.

The character recognition block 40 'includes a feature extraction unit 41, a dictionary unit 42, an identification unit 43, and a predicted shape information unit 4.
4. Image correction processing unit 45 ', corrected image determination unit 4
6 ′ and a rejection determination unit 47, perform character extraction, image correction processing, etc., and then perform character recognition.

Here, in the character recognition block 40 ', the feature extraction unit 41, the dictionary unit 42, the identification unit 43, and the predicted shape information unit 4
The configuration, function, and operation of 4 and rejection determination unit 47 are the same as those of character recognition unit 40 of character recognition device 100 in FIG. Further, the image correction processing unit 45 ′ includes the character code Co1 and the distance Di1 of the candidate character obtained by the identification unit 43.
Based on the ruled line contact information fi stored in the ruled line contact information storage unit 30, all the operations for correcting the cut-out character image Ci1 so as to approximate the predicted shape according to a predetermined priority order are performed. Repeat about the shape,
A group of corrected images Ci2 is obtained. Note that the character cutout unit 2
If a character image Ci1 cut out at 0 is corrected several times in accordance with the priority and there is no image to be corrected any more, a reject code (rejection code) is output. The priority here is a correction order when the cut-out character image is corrected. For example, when correcting an image that is in contact with the upper, middle, and lower ruled lines, the order in which to correct first is determined. (For example,
(Up → middle → bottom or bottom → middle → top), which is a part of the prediction information.

The corrected image determination section 46 'determines the reliability of the corrected image Ci2 corrected by the image correction processing section 45' as an image. The image is output to the rejection determination unit 47 as the image Ci3, and if not, the corrected image Ci2 is given as a cutout character to the image correction processing unit 45 'so that the image correction is performed again by the image correction processing unit 45'. FIG. 8 shows the character recognition device 1 of FIG.
12 is a flowchart showing an example of a character recognition operation by 00 ', which is provided with a correction image creation step (S6') for correcting a cut-out character image in accordance with a priority order and a step (S7 ') for determining the reliability of the correction image. In this example, when the reliability of the corrected image is low, a new character image can be obtained by repeating the generation of the correction again. The control of the operation sequence of each step is performed by the control unit. Steps S1 to S5 in FIG.
(Removal of ruled lines, etc.-identification processing of cut-out characters) and S8-
The operation up to S10 (rejection determination to code output) is the same as the character recognition operation in FIG. Hereinafter, the operations of S6 ′ and S7 ′ will be described.

Step S6 ′: (Acquisition of Corrected Image Group) In FIG. 8, the image correction processing unit 45 ′ corresponds to the character image Ci1 cut out by the character cutout unit 20 and the conditions obtained by the above operation ( Based on the ruled line contact information If2 of the cut-out character image (for example, the character image 42, the recognition candidate character candidate Co1 and the distance calculation result Di1)), the shape information that satisfies the conditions acquired by the above operation from the predicted shape information unit 44 is obtained. The operation of correcting the cut-out character image Ci1 obtained in accordance with the predetermined priority order so as to approach the predicted shape in accordance with the predetermined priority order is repeated for all the predicted shapes, and the corrected image Ci2 is obtained.
The group is obtained, and the process proceeds to S7 '. Note that the correction image Ci2
If not, the process proceeds to S10 (reject code output process).

Step S7 ': (Determining the reliability of the corrected image) The corrected image determining unit 46' includes an image correction processing unit 45 '.
The reliability as an image is determined for each corrected image in the group of corrected images Ci2 corrected by
The correction image having the highest reliability among the correction images determined to have reliability is output to the rejection determination unit 47 as the recognition candidate character image Ci3, and the process proceeds to S8. Otherwise, the image correction processing unit 45 is performed again. In order to perform the image correction by ', the corrected image Ci2 is given as a cutout character to the image correction processing unit 45', and the process returns to S6 '. As a method of determining the reliability of the group of corrected images Ci2, in the embodiment, the feature amount of each corrected image Ci2 is calculated, and the distance calculation from the standard feature amount Fdd stored in each template of the dictionary unit 42 is performed. Is performed, and based on the result, the distance r ′ of each corrected image Ci2 is smaller than the threshold δ ′ (δ ′> δ) (that is, r′−δ ′ <
0) A corrected image Ci3 having a distance from another category is extracted, and a corrected image in which the character code Co1 of the upper candidate character has the largest number of matches is extracted from the corrected images, and a recognized candidate character image is extracted. And Note that the reliability determination method is not limited to this method. Hereinafter, a specific operation example of character recognition by the character recognition device 100 ′ of FIG.
5, S8 to S10 will be described based on the flowchart of FIG. 2).

The read image Im1 is the image shown in FIG. 3, and the image Im2 from which the ruled lines and the like have been removed by the ruled line removing unit 10 is the image shown in FIG. 4 (S1).
Here, a character image Ci1 for two characters is cut out by the character cutout unit 20 (S2). In the following, the lower one part of the cut-out character image of two characters is indicated by a frame 3
Image 41 that has lost information at the bottom when it came into contact with 2
(The character "2" (FIG. 9A) is taken as an example.

In the cut-out image 41 shown in FIG. 9A cut out by the character cut-out unit 20, the position of the portion (81) in contact with the lower side of the frame line 32, the number of contact points (one place), and the ruled line Is stored in the ruled line contact information storage unit 30 as ruled line contact information If2 (S3).

Next, the feature extraction unit 41 performs feature extraction on the cut-out image 41 to calculate a feature amount Fd1 (S4), and the identification unit 43 compares the feature amount Fd1 with the dictionary unit 14.
2 is a standard feature F stored in each template
dd, and obtains the recognition candidate character codes Co1 and the distance calculation result Di1 from the smallest distance (the order of the closest feature) to the top three as the recognition candidate information (S
5). For the sake of explanation, the recognition candidate character codes Co1 up to the third place obtained here are character codes in which the first place recognition character code and the second place recognition character code indicate "7".
The position recognition character code is a character code indicating “9”.

Next, in the image correction processing section 45 ', the recognition candidate information (the recognition candidate character code Co1 and the distance calculation result Di) obtained in step S5 from the predicted shape information section 44 are obtained.
1) and the line contact information I of this cut-out character image (FIG. 9A) stored in the line contact information storage unit 30.
The predicted shape information that satisfies f2 is acquired, and the operation of correcting the cut-out character image so as to approach the predicted shape according to a predetermined priority is repeated for all the predicted shapes to obtain a corrected image group (cut-out image group). In the case of 41, in the first cycle (cycles of S6 'and S7'), it is determined that there is a possibility of "7" from the predicted shape information, and the corrected image Ci2 of FIG. 9B is obtained (S6 ').

Next, the correction image determination section 46 'determines the reliability of the correction image "7" obtained in step S6'. Here, if the distance between the character image of “7” in FIG. 9B and the standard feature Fdd of the dictionary unit 42 is large (r′−δ ′> 0), the corrected image determination unit 46 'Moves to S6' in order to judge that the reliability as an image is low and perform image judgment again (S6).
7 ').

The image correction processing unit 45 'predicts an operation of correcting the cut-out character image 41 obtained in step S2 so that the cut-out character image 41 approaches the shape to be predicted according to the following priority order. Repeatedly for all the shapes, a new correction image (Fig. 9
(A correction image of “2” shown in FIG. 3C) is acquired (S
6 ').

Again, at S7 ', the corrected image determination unit 46'
Then, the reliability of the corrected image “2” obtained in step S6 ′ is determined. Here, assuming that the distance between the character image of “2” in FIG. 9C and the standard feature Fdd held in the dictionary unit 42 is small, the character image of “2” becomes the second image (as described above). Since the character code Co1 matches as the first candidate and the second candidate, and the top candidate has the largest number of matches,
The corrected image determination unit 46 'determines that the reliability as an image is high, and shifts to S8 for rejection determination.

In step S8 (FIG. 2), rejection determination section 47
Calculates the distance between the corrected image Ci3 obtained in step S7 ′ and the standard feature value Fdd stored in each template of the dictionary unit 42, and compares the distance r with a predetermined threshold δ. (In this example, if the distance r−δ <0 (that is, the distance <threshold) of the recognition candidate (corrected image Ci3) “2” is satisfied (ie, the distance <threshold value), the process proceeds to S9 where the character code of the character “2” is output. To S10, the reject code is output.

With the above configuration, as shown in FIG. 9A, the corrected image reproducing the lost character image as shown in FIG. 10C even if the character loses the information of the true image as shown in FIG. Since it can be obtained, a highly reliable recognition result can be obtained.

Although several embodiments of the present invention have been described above, the present invention is not limited to these embodiments, and it goes without saying that various modifications can be made.

[0057]

As described above, according to the character recognition methods of the first to sixth inventions and the character recognition device of the fourth invention, when removing the ruled line, the information of the portion in contact with the ruled line is removed. Since the character image is corrected using the information stored, the character image can be corrected irrespective of the stroke direction, and the character touching the ruled line is separated into several parts (blocks). Since the correction can be performed even if the correction is performed, character recognition with a high recognition rate can be realized even if a form or the like in which ruled lines of a non-dropout color or the like are printed in one color is used.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a configuration of an embodiment of a character recognition device of the present invention.

FIG. 2 is a flowchart illustrating an example of a character recognition operation performed by the character recognition device of FIG. 1;

FIG. 3 is a diagram illustrating an example of a read character image.

FIG. 4 is a diagram showing an example of a character image after removing ruled lines and the like.

FIG. 5 is a diagram showing an example of a character image after one character is cut out.

FIG. 6 is a diagram illustrating a result of image correction and recognition using a cut-out character as an example.

FIG. 7 is a block diagram showing a configuration of an embodiment of the character recognition device of the present invention.

8 is a flowchart illustrating an example of a character recognition operation performed by the character recognition device of FIG. 7;

FIG. 9 is a diagram illustrating a result of image correction and recognition using a cut-out character as an example.

FIG. 10 is a diagram illustrating a comparison between a recognition result obtained by the character recognition method of the present invention and a recognition result obtained by the conventional character recognition method.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 10 Ruled line removal part (ruled line removal means) 20 Character cutout part (cutout means, ruled line contact information acquisition means) 41 Feature extraction part (character recognition means) 42 Dictionary part (character recognition means) 43 Identification part (character recognition means) 45, 45 'image correction processing section (image correction means) 46 corrected image selection section (correction image selection means) 46' correction image determination section (correction image determination means) 47 rejection determination section (recognized character output evaluation means) 100, 100 'characters Recognition device

Claims (6)

[Claims] 1. A character recognition method for removing characters such as ruled lines from a read image of a read original and extracting a character image one character at a time to perform character recognition, wherein the character extracted when the character image is extracted. Obtain contact information between the image and the ruled line, etc., perform recognition processing of the cut-out character, and, based on the recognition result obtained in the recognition processing and the contact information, register the cut-out character image in advance. Obtaining a corrected image group corrected to approximate the shape predicted from the predicted shape information obtained, selecting a recognition candidate image from the corrected image group, and determining whether to output a recognition character corresponding to the recognition candidate image. A character recognition method characterized by evaluating whether or not a character is recognized. 2. A character recognition method for removing characters such as ruled lines from a read image of a read document and extracting a character image one character at a time to perform character recognition, wherein the character extracted when the character image is extracted. Obtain contact information between the image and the ruled line, etc., perform recognition processing of the cut-out character, and, based on the recognition result obtained in the recognition processing and the contact information, register the cut-out character image in advance. A corrected image group corrected to approximate the shape predicted from the predicted shape information obtained is obtained, it is checked whether the corrected image in the corrected image group satisfies a predetermined condition, and the predetermined condition is satisfied. Extracting a recognition candidate image from the corrected image and evaluating whether to output a recognition character corresponding to the recognition candidate image or not. Character recognition method. 3. When all of the corrected images in the corrected image group do not satisfy the predetermined condition, further based on the recognition result of the character image and the contact information, According to the predetermined correction priority, a corrected image group corrected so as to approach the shape predicted from the pre-registered predicted shape information is obtained, and further, whether the corrected image in the corrected image group satisfies predetermined conditions. 3. The character recognition method according to claim 2, wherein an operation of extracting a recognition candidate image from the corrected images satisfying the predetermined condition is checked. 4. A ruled line removing means for removing an image such as a ruled line from a read image of a read document; a cutout means for cutting out a character image one by one from a character image from which a ruled line or the like has been removed by the ruled line removing means; Ruled line contact information acquiring means for acquiring contact information of a portion where the ruled line or the like was in contact from the character image from which the ruled line or the like has been removed by the removing means; and recognizing means for recognizing the character image cut out by the cutout means. Based on the recognition result and the contact information obtained by the recognition processing by the recognition unit, based on the contact information, a corrected image group corrected so that the cut-out character image approaches a shape predicted from pre-registered predicted shape information Image correction means for obtaining, and a corrected image obtained by the image correction means Correction image selection means for selecting a recognition candidate image from among the above, recognition character output evaluation means for evaluating whether to output a recognition character corresponding to the recognition candidate image obtained by the correction image selection means, A character recognition device comprising: 5. A ruled line removing means for removing an image such as a ruled line from a read image of a read original; a cutout means for cutting out a character image one by one from a character image from which a ruled line or the like has been removed by the ruled line removing means; Ruled line contact information acquiring means for acquiring contact information of a portion where the ruled line or the like was in contact from the character image from which the ruled line or the like has been removed by the removing means; and recognizing means for recognizing the character image cut out by the cutout means. Based on the recognition result and the contact information obtained by the recognition processing by the recognition unit, based on the contact information, a corrected image group corrected so that the cut-out character image approaches a shape predicted from pre-registered predicted shape information Image correction means for obtaining, and a character image corrected by the character image correction means. Determining whether a predetermined condition is satisfied, and extracting a recognition candidate image from among the corrected images satisfying the predetermined condition; and a recognition candidate image obtained by the corrected image determination unit. A recognition character output evaluating means for evaluating whether or not to output a corresponding recognition character; 6. The image correction means, if all the corrected images do not satisfy a predetermined condition as a result of the determination by the corrected image determination means, further based on the recognition result of the character image and the contact information. 6. The character recognition device according to claim 5, further comprising: means for correcting the cut-out character image so as to approach a shape predicted from pre-registered predicted shape information according to a predetermined correction priority. apparatus.