JPH05128299A - Character recognition method and device therefor - Google Patents

Abstract

(57) [Summary] [Purpose] By changing the method of character segmentation and retrying character segmentation, erroneous recognition due to character segmentation is reduced. [Structure] After inputting an image original, characters are cut out by projection in a character cutting unit 109, and character image blocks 601-
We get 606. After that, for the character image blocks whose similarity is calculated by the similarity calculator 110 and the similarity is lower than a predetermined threshold value (TH), the method of combining the character image blocks is changed and the By cutting out, character image blocks 607 to 608 different from the first one are obtained.
The similarity calculation is executed again for this character image block, and the result is evaluated again. In this way, the method of combining the character image blocks is changed, the similarity calculation is performed, and the result is evaluated again to obtain a character image block cut out with higher accuracy.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character recognition method and apparatus.

[0002]

2. Description of the Related Art A conventional character recognition device inputs an image original (step S201) and cuts out characters one by one from the input image (step S202) in accordance with the procedure of the flowchart shown in FIG. 8 to extract character features. After that (step S203), the degree of similarity with the previously-obtained characteristics of each category is calculated (step S204), and the characters belonging to the category with the closest degree of similarity are used as the recognition result (step S2).
05).

[0003]

However, in the above-mentioned conventional example, when the character is erroneously cut out, even if the character of the category obtained from the similarity calculation has the highest similarity, the value of the similarity is There was a case where the recognition result was suspicious because it was not so big and it was empirical. This was an erroneous recognition that occurs because the erroneous character segmentation is not corrected and the character recognition process is executed as it is.

The present invention has been made in view of the above-mentioned conventional example, and an object of the present invention is to provide a character recognition method and an apparatus therefor capable of correcting a character cutout error.

[0005]

In order to achieve the above object, the character recognition method of the present invention comprises the following steps. That is, a character recognition method for recognizing a character drawn on an image manuscript by inputting an image manuscript, wherein an input step of inputting the image manuscript and a character drawn on the image manuscript input by the input step. And a character recognition step of extracting the features of the characters cut out by the cutting step, comparing the information with the information stored in a predetermined character dictionary to obtain the similarity, and performing character recognition. , A control step of evaluating the result of the character recognition by the recognition step, and based on the evaluation, controlling the character separation to be different from the previous one and returning to the cutout step to perform the character cutout again. It has a character recognition method.

According to another aspect of the present invention, there is provided a character recognition device for inputting an image original and recognizing characters drawn on the image original, the input means inputting the image original, and the input means for inputting the image original. A cut-out means for separating and cutting out the characters drawn on the image original one by one, and a characteristic of the character cut out by the cut-out means,
A recognition unit that performs character recognition by obtaining similarity by comparing with information stored in a predetermined character dictionary, evaluates the result of the character recognition by the recognition unit, and based on the evaluation,
The character recognition device is characterized by further comprising control means for controlling character segmentation so that the character segmentation is performed differently from the previous time and the character segmentation is performed again.

[0007]

With the above configuration, the present invention performs character recognition of a character cut out from an input image, evaluates the result of the character recognition, and again cuts out the character by differentiating the character separation from the previous time. Operate to control to do.

[0008]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENT A preferred embodiment of the present invention will be described in detail below with reference to the accompanying drawings.

[General Description of Apparatus (FIG. 1)] FIG. 1 is a block diagram showing the configuration of a character recognition apparatus which is a typical embodiment of the present invention. In FIG. 1, 101 is a CPU which controls the entire apparatus and controls execution of character recognition processing, 102 is a keyboard (KB) for inputting and correcting characters, and 103.
Is a pointing device (PD) such as a mouse, 104
Is an image scanner (SCAN) that reads images, 10
5 is an image scanner interface, 106 is a CP
RO for storing control programs executed by U101
RAs for storing recognition results, character cutting information, and the like.
M and 108 are displays for displaying recognition results and the like. Reference numeral 109 is a character cutout unit that cuts out characters one by one from the character image, and 110 is a similarity calculation unit that extracts features from the cutout character image and calculates the similarity to each category.

In the RAM 107, a character image storage area 111 for storing a character image read by the image scanner 104, a character cutting information storage area 112 for storing the output of the character cutting section 109, and a similarity calculation section 110. A recognition result storage area 113 for storing the output is secured.

[Description of Character Recognition Processing (FIGS. 2 to 4)] Next, the character recognition processing executed by using the character recognition device having the above-described configuration will be described with reference to the flowchart shown in FIG.

First, in step S101, a character image is input from the image scanner 104, and the character image storage area 1
It is stored in 11. In step S102, the character cutout unit 109 cuts out characters one by one from the stored character image. The method of cutting out is not limited, but in the present embodiment, a method of projecting a character image is used.

FIG. 3 is a diagram showing how character segmentation is performed according to a method for projecting a character image. In FIG. 3, 401 is an input document image and 402 is a projection plane. This projection is taken in the horizontal direction if the image original to be input is a horizontal original, and in the vertical direction if it is a vertical original. In the case of FIG. 3, since it is a horizontal document, a projection 404 is taken for the character image 403, and a line rectangle 405 is obtained. FIG. 4 is a diagram showing a state in which a vertical projection is taken on the obtained row rectangle 405. If a projection in the horizontal direction is obtained, then the projection plane is rotated 90 degrees and the projection is taken to form a character rectangle 50.
1 is obtained and the characters are cut out one by one. The result of this character segmentation is output to the character segmentation information storage area 112.

Next, in step S103, a feature is extracted from the character image of the determined character rectangle 501, and then in step S104, the similarity calculating unit 11 uses the extracted feature.
The similarity calculation is performed with 0, and the calculation result is output to the recognition result storage area 113. The output information includes, for example, n calculated values and character codes from a category having a high degree of similarity.

The processing of steps S101 to S104 is based on a known technique.

Now, in step S105, step S1
It is checked how many times the series of processing from 02 to S104 is executed. Here, if it is determined that the series of processes has been performed for the first time, the process proceeds to step S106, and if it is determined that the process has been performed twice or more, the process proceeds to step S107. In step S106, the character image block for which character cutting is to be performed again is selected. This selection is performed when the calculated value of the category for which the maximum similarity is obtained is smaller than a predetermined threshold value (TH). For the character image block selected here, "1 (on state)" is set to the re-character cut-out status flag (IFLG) managed by the CPU 101. On the other hand, in step S107, the character cutout status flag (IFLG) of the character for which the character cutout has been executed again is checked, and the value of the similarity is a predetermined threshold value (TH).
The character cutout status flag (IFLG) is set to "0" for a character that is sufficiently higher and does not need to be recut out.

This character cutout status flag (IF
LG) is a flag that is set for each of the cut characters and is stored in the character cutting information storage area 112 of the RAM 107.

Next, in step S108, it is checked whether or not there is a character whose character cutout status flag (IFLG) is on. Here, if there is a character in the ON state, the process returns to step S102 and the character cutting is performed again for the character. On the other hand, if there is no character in the ON state, the process proceeds to step S109, the result is displayed on the display 108, and the process ends.

[Specific Example of Character Recognition Processing (FIG. 5)] The above processing will be specifically described on the assumption that the character image shown in FIG. 5 is input. Here, the characters shown in FIG. 5A are input image characters.

In the first character segmentation, as described in step S102, first, projection is performed to extract six image blocks 60 from the input image character as shown in FIG. 5B.
1 to 606 are obtained. Next, a narrow image block (here, 601-6
03) are combined to form four character image blocks 605 to 607 as shown in FIG. Here, the standard character width is an average value of the widths of character rectangles having a width close to the height of the line rectangle 405.

After that, after the processing of steps S103 to S105, a character image block to be subjected to character segmentation is selected again in step S106. Here, it is assumed that the character cutout status flag (IFLG) of the character image blocks 607 to 608 shown in FIG. 5C is on. In the character extraction in step S202 from the second time onward, only the character image block for which the character extraction status flag (IFLG) is ON is the processing target.

In the second character segmentation, first, the combination of the head of consecutive re-character segmented image blocks and the blocks after it is examined. In the case of FIG. 5B, the character image blocks 601 to 604 are continuous, and the first 601 is 6
Check if it binds to 02-604. However,
The width of the combined character image block is set to be within a constant multiple of the standard character width, and the number of combinations is limited.
As a result, it is assumed that there are three patterns of combination, that is, only 601, a combination of 601 and 602, and 601 to 603. The processes of steps S103 to S105 are executed for these combined character image blocks, and finally, the combination having the highest similarity and sufficiently higher than the predetermined threshold value (TH) in step S107 is adopted as the cutout result. Here, in the case of only the character image block 601, if the condition is satisfied, the character cutout status flag (IFLG) of the character image block 601 is turned off. At this point, the character segmentation is in the state shown in FIG.

In the next third character segmentation, the character image block 602 and other blocks (here, 603 and 60) are extracted.
4) Check the binding with. If it is determined in the third trial that the similarity between the character image blocks 602 and 603 is maximized and that the similarity is sufficiently higher than a predetermined threshold value (TH), the two are combined into 1 One character image block 609 is set, and the character cutout status flag (IFLG) for this block is turned off.

The character image block 604 remains in the final fourth character cut-out, and the processes of steps S102 to S105 are executed for this block, and step S
In 107, a character cutout status flag (IF
LG) is turned off. By the above processing, FIG.
A divided character image block 60 as shown in (e)
1, 609 and 604-606 are obtained.

Here, if the character cut-out status flag (IFLG) is turned on in step S106, the character will not be divided into a plurality of character image blocks in the next character cut-out process, and the adjacent characters will not be divided. If the image block of No. does not exist, it is determined in step S106 of the subsequent loop that there is no error in the character cutting processing, and the character cutting status flag (IFLG) is turned off so that the character cutting is not performed again.

Therefore, according to the present embodiment, first, the entire character is cut out from the input image, the feature extraction / similarity calculation is performed, and the portion that needs to be extracted again is determined based on the calculation result of the similarity calculation. After that, by repeating the character segmentation, the feature extraction, and the similarity calculation, it is possible to correct the erroneous recognition caused by the character segmentation even when the series of recognition processes is completed.

[0027]

[Other Embodiments] Here, with respect to the process of successively performing character recognition while cutting out characters one by one, which is executed by the above-described character recognition device as a common device, the flowchart shown in FIG. 6 and the specific example shown in FIG. Explain. Since the apparatus is the same as that of the above-described embodiment, a description thereof will be omitted by using a common apparatus reference number, and in the processing of the present embodiment, common steps with the above-described embodiment will be denoted by common step reference numbers. And the description is omitted.

In the present embodiment, a logical marker is used in order to grasp the execution status of the character recognition processing for each character. This marker is always located in the character image block where the character is cut out, and the character is cut out based on the marker position. Then, when it is determined that the character cutout is satisfactory, the marker is moved to the next character image block, and the next character image block is cut out.

After the input of the image original in step S101, in step S102, only the first character image of the input image is cut out from the input image as shown in FIG. 7A. As a result, the character image block 801 is cut out from the input image. At this point, the marker 70 is located on the left side of the character image block 801. Here, similarly to the above-described embodiment, the line rectangle 405 is first obtained in accordance with the method of character segmentation by projection, and the character image for one character is segmented by projecting from the head portion of the head line in the row direction and the vertical direction. ..

Next, it is checked whether or not there are characters cut out in step S701. If it is determined that there is a clipped character image block, the process proceeds to step S
Going to 103 to S104, the feature extraction and the similarity calculation are executed. On the other hand, if it is determined that there is no character image block to be cut out, the process proceeds to step S109, the process result is displayed, and the process ends.

After the processing in steps S103 to S104, it is determined in step S702 whether or not to execute the character cutting processing again based on the calculated value obtained from the similarity calculation. Here, if the calculated value of the category for which the degree of similarity is obtained is smaller than the predetermined threshold value (TH) as in the above-described embodiment, the process proceeds to step S703, and it is considered that the character cutout process is executed again. For this purpose, the marker 70
Does not move. On the other hand, if the calculated value is larger than the predetermined threshold value (TH), the process proceeds to step S704, where it is determined to proceed to the cutting out of the next character. Therefore, the marker 70 moves to the left side of the next character image block or the right side of the determined character image block. Step S70
3 and S704, the process returns to step S102 to continue the character segmentation.

Now, in step S702, step S
Since it is positioned between the loop processing of 102 to step S703 or step S704, it is determined whether or not the character cutting processing is to be executed again and again. Therefore, if the calculated value of the similarity calculation of the previous time is compared with the calculated value of the similarity between this time and the previous character cutout shows a relatively better result than the result of the current character cutout, the character cutout is performed again by the previous judgment. There is a case in which it is determined that the character is to be executed, but it is determined in this determination that no more character cutout will be performed, and that the previous character cutout result is adopted.

In order to specifically take this into consideration, description will be given with reference to FIG. 7, assuming that the image original "(various)" is input.

When the character image block 801 is taken out as shown in FIG. 7A in the first character cutout and it is judged in step S702 to proceed to step S703, the second character cutout is performed as shown in FIG. The character image block 8 combined with the next character image block as shown in b).
It is assumed that 02 is taken out. At this time, the marker 70 indicating the base point of the character cutout is at the left end of the character image block 801 or 802, as shown in FIGS.

However, if it is determined in the next step S702 that the character image block 801 should be cut out, the process proceeds to step S704, where the marker 70 is moved to the right side of the character image block 801, and the third time. In the character cutout, a character image block 803 is taken out as shown in FIG. At this time, the marker 70 indicating the base point of the character cutout moves to the left end of the character image block 803, as shown in FIG. 7C. When the process proceeds to step S703 by the determination in step S702 for the third time, the position of the marker 70 does not move, and in the fourth character cutout, the marker image is combined with the next character image block as shown in FIG. 7D. The character image block 804 is fetched. When it is determined that the character image block 804 is sufficient for the character segmentation in the fourth determination in step S702,
The process proceeds to step S704, the marker 70 is advanced to the right of the character image block 804, and the fifth character segmentation is performed.
The next character image block is cut out as shown in FIG.

Similarly, the character image blocks are sequentially cut out, and when all the character images are cut out, the process proceeds to step S109 to display the result.

Therefore, according to this embodiment, the character recognition process is performed while preventing the erroneous recognition due to the character cut-out by sequentially advancing the character recognition process of the character cut-out and the feature extraction / similarity calculation for each character. It is possible to

Although the above embodiment has been described by using the device having the configuration in which the processing of the character cutout unit and the similarity calculation unit is executed by the dedicated processing processor, the present invention is not limited to this. For example, the CPU may be a device configured to execute these processes.

Further, the character cutting method has been described by adopting the projection method in the above embodiment, but the present invention is not limited to this. For example, another method such as a boundary line tracing method may be used. You can also

Furthermore, in the above-mentioned embodiments, the relatively short character "(various characters) is used as the input image and the character segmentation is executed for all the input images. However, the present invention is not limited to this. For example, with respect to an input image such as a long sentence, it is possible to execute character cutting out within a specified range or line by line.

The present invention may be applied to a system composed of a plurality of devices or an apparatus composed of one device. Further, it goes without saying that the present invention can be applied to the case where it is achieved by supplying a program to a system or an apparatus.

[0042]

As described above, according to the present invention, after character recognition is performed on a character cut out by character cutting, the result of character recognition can be evaluated and the character cutting can be tried again. This has the effect of reducing erroneous recognition due to clipping. As a result, erroneous recognition is reduced, and there is also an advantage that erroneous recognition correction work is reduced and document input time is shortened.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of a character recognition device that is a typical embodiment of the present invention.

FIG. 2 is a flowchart showing a character recognition process capable of re-execution of a character segmentation executed by the character recognition device shown in FIG.

FIG. 3 is a diagram showing a state in which a horizontally projected character is projected in a horizontal direction.

FIG. 4 is a diagram illustrating a state in which a horizontally projected character is projected in a vertical direction.

FIG. 5 is a diagram for explaining character cutout.

FIG. 6 is a flowchart showing a character recognition process according to another embodiment.

FIG. 7 is a diagram showing how character cutout is corrected according to another embodiment.

FIG. 8 is a flowchart showing a character recognition process according to a conventional example.

[Explanation of symbols]

 101 CPU 102 Keyboard 103 Pointing Device 104 Scanner 105 Interface 106 ROM 107 RAM 108 Display 109 Character Extraction Unit 110 Similarity Calculation Unit

Claims (4)

[Claims] 1. A character recognition method for recognizing a character drawn on an image original by inputting the image original, comprising an input step of inputting the image original, and a drawing on the image original input by the input step. A character cutting process is performed to separate the extracted characters one by one, and the characteristics of the characters cut out by the cutting process are extracted and compared with the information stored in a predetermined character dictionary to obtain the similarity and perform character recognition. A recognition step; and a control step of evaluating the result of the character recognition by the recognition step, and controlling the character separation to be different from the previous time and return to the cutout step to perform the character cutout again based on the evaluation. Character recognition method characterized by the following. 2. A character recognition device for inputting an image original and recognizing characters drawn on the image original, comprising: input means for inputting the image original; and drawing on the image original input by the input means. Cutout means for separating the extracted characters one by one and the characteristics of the characters cut out by the cutout means are extracted and compared with the information stored in a predetermined character dictionary to obtain similarity and perform character recognition. A recognition unit; and a control unit that evaluates the result of the character recognition by the recognition unit and controls the character separation to be different from the previous one so that the character cutout is performed again based on the evaluation. A character recognition device characterized by the above. 3. The processing of the recognizing means and the control means is performed each time the characters drawn on the image original are separated and cut out one by one by a predetermined length. The character recognition device according to item. 4. The character according to claim 2, wherein the processing of the recognition means and the control means is performed every time the character drawn on the image original is separated and cut out one by one. Recognition device.