JPH1125221A - Optical character reader read - Google Patents

Abstract

PROBLEM TO BE SOLVED: To adjust a binary picture to a picture suited to character recognition in an optical character reader for operating character recognition by turning the picture of a slip into a binary picture. SOLUTION: A binarizing circuit 13 generates a binary picture by binarizing the multilevel picture of a slip stored in a multilevel data buffer 12 by a set threshold, and a filter part 15 shapes a character pattern in the binary picture by a filter processing, and stores it in a binary data buffer 16-1. A recognizing circuit 16-2 operates character recognition for the binary picture stored in the binary data buffer 16-1, and in case of illegibility in the character recognition, the recognizing circuit 16-2 operates the character recognition again by operating binarization by changing a threshold in the binarizing circuit 13. At this point, a filter selecting part 14 selects a plurality of filters 15-1-15-n in the filter part 15 according to the number of times of the character recognition, and the adjustment of the binary picture is operated in a shaping manner different from that in the previous filter processing.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an optical character reader (hereinafter, referred to as OCR) for optically acquiring an image of a form and recognizing characters described on the form.

[0002]

2. Description of the Related Art FIG. 2 is a configuration diagram showing a main part of a conventional OCR. The OCR includes a photoelectric conversion unit 1 that optically acquires a form image and generates multivalued data corresponding to the acquired image by photoelectric conversion, and a multivalued data buffer 2 that stores the multivalued data. Have. On the output side of the multi-level data buffer 2, a binarizing circuit 3a and a filter 3
b are sequentially connected, and a recognition unit 4 is arranged on the output side of the filter 3b. The recognizing unit 4 includes a binarized data buffer 4a connected to an output terminal of the filter 3b,
It comprises a recognition circuit 4b connected to the output side of the quantified data buffer 4b. The OCR is further provided with a control unit 5. The control unit 5 includes the photoelectric conversion unit 1,
The multi-level data buffer 2 and the recognition circuit 4b are controlled. Next, the operation of each unit in the OCR will be described. The photoelectric conversion unit 1 converts a graphic character (hereinafter, simply referred to as a character) in a character portion on the form into digital data of multi-value gradation, and this digital data is provided to a multi-value data buffer 2. The multi-level data buffer 2 stores multi-level digital data and stores a multi-level image of a form.

[0003] The binarizing circuit 3a includes a multi-level data buffer 2
Of the multi-valued digital data read from the document, and generates binary data to be a binary image of the form. Filter 3b
Smoothes the binary data by filter processing, and
Shape the value image. The filtered binary data is
The binary data is supplied to the binary data buffer 4a in the recognizing unit 4 and accumulated. The recognition circuit 4b performs a recognition process on the binary image stored in the binary data buffer 4a, and recognizes a character described in a form. The recognition result is output to the control unit 5. In the above operations, sending binary data from the binary data buffer 4a to the recognition circuit 4b so that highly accurate recognition can be performed using the recognition result and information on the shape of the character as parameters is called recognition control. . The flow of the control will be described with reference to FIG. FIG. 3 is a flowchart showing the recognition control of FIG. 2, and shows the processes S1 to S7. In the process S1 of FIG. 3, the characters entered on the form are converted into multi-valued image data (multi-valued gradation digital data) by the operation of the photoelectric conversion unit 1 using a sensor system (not shown). The multi-valued image data is sliced with a fixed threshold value and converted into binary data in the binarization process of the process S2. In step S3, an average line width W for evaluating the line width of the character is calculated from the binary data.

FIG. 4 is an explanatory diagram for calculating an average line width.
Since the actual character pattern is complicated, an example in which the average line width W is calculated using four black points using a simplified model of a rectangle as shown in FIG. 4 will be described. For example, if character pattern 6 is 4
Assuming that it is formed of × 10 black spots, the total number of black spots is 40. This character pattern 6 has a size of 2 ×
Scanning is performed using the pattern 7 of the grid 2. At this time, one of the 2 × 2 grid patterns 7 is set as a reference black point 8, and the pixels of the character pattern 6 are scanned with the reference black point 8. In the scan, 2
When all the pixels of the character pattern 6 corresponding to the four points of the × 2 lattice pattern 7 are black, the points on the character pattern 6 corresponding to the reference black point 8 at this time are defined as four black points. In this way, when the pixels of the character pattern 6 are scanned, the pixels that become four black dots become the area 9 indicated by hatching in FIG. Based on this area 9, the average line width W is obtained by the following equation (1). W = A / (A−Q) (1) where A; total number of black spots Q; four black spots When this equation (1) is applied to FIG. 4, the average line width W is 40 /
(40−27) = 3.1.

FIG. 5 is an explanatory diagram of a change in a character pattern according to a binarization threshold. Digital data of multi-level gradation in the OCR is generally composed of eight gradations, and the binarization threshold TH at the time of slicing in the binarization process takes a value from 0 to 7. 0 is the lightest and 7 is the darkest gradation. Threshold T
By changing H from 0 to 7, the character pattern is changed and shaped as shown in FIG. As the threshold value TH increases, the line width of the character pattern increases. When the threshold value TH is 0, the character pattern is blurred, and the pattern is cut. Usually, the line width is extremely thick,
If it is fine, the characters will be crushed, faded, and cut, which is not suitable for character recognition. In the conventional recognition control, the line width is used as a parameter, and recognition is performed when the line width takes a value within a certain range. At the first binarization stage after the unknown character pattern is read by the sensor system in the photoelectric conversion unit 1, it is not possible to determine which of the thresholds TH to be binarized from 0 to 7. Therefore, the current OCR
Sets the first threshold value TH as a fixed value. This value is called an inrush threshold.

In a process S3 shown in FIG. 3, an average line width W is calculated based on the data binarized by the inrush threshold value.
In processing S4, when the calculated average line width W is within the allowable line width range, it is determined that the pattern is recognizable (Y),
The data is sent to the recognition unit 4. If the average line width W of the character pattern does not fall within this range, it is determined that it is inappropriate for recognition (N), the binarization threshold value TH is changed in step S5, and steps S3, S4, and S5 are performed again. . That is, when it is determined that the average line width W is not within the range of the line width allowable value, the processes S3 and S3 are performed.
The operations of 4, S5 are repeated. At this time, when the average line width W is on the thick side, the threshold value TH is decreased by 1, and when the average line width W is on the thin side, the threshold value TH is increased by 1. In the process S6 when the average line width W is within the allowable line width range, the recognition circuit 4b performs recognition using a recognition dictionary provided with an algorithm for determining the category of each character. Finally, the recognition result obtained in step S6 is displayed on a display or the like (not shown).

[0007]

However, the conventional OCR has the following problems. FIG. 6 (a),
2B is an explanatory diagram of the problem in FIG. 2, and FIG. 2A is a diagram when the character pattern changes according to the threshold value TH.
Indicates a case where there is no change. Depending on the type of the form, adjustment of the sensor in the photoelectric conversion unit 1, and the like, the binarized character pattern may hardly change for each threshold value TH. Normally, as shown in FIG. 6A, for example, if the threshold value TH is changed from 4 to 6, the character pattern should also change in accordance with the change. However, for example, FIG. May hardly change. If such a character pattern is unread in the first pattern, it will be unread in the second and subsequent times, and the current OCR
According to the specification, when the number of scans is exceeded, the final result is unreadable characters. That is, there is a problem that the effect of the recognition control is hard to appear, and as a result, the recognition accuracy of the OCR is deteriorated.

[0008]

In order to solve the above-mentioned problems, the first to eleventh inventions optically capture an image of a form in which characters are written, photoelectrically convert the image, and convert the image into an image. A photoelectric conversion unit for generating multi-value data representing an image in multi-valued gradation; a multi-value data buffer for storing the multi-value data; and a multi-value data read from the multi-value data buffer at a set threshold of 2 A binarization circuit for generating a binary image corresponding to the form, and filtering the binary image output from the binarization circuit to shape a character pattern in the binary image A filter unit, and a recognition circuit that recognizes the character by performing a recognition process on the filtered binary image, and when the recognition circuit cannot recognize the character, the threshold value is changed. Binarization and re-filling In recognizing OCR the characters in re-recognition processing by the recognition unit after performing the process, and the following structure.

That is, the filter section has a plurality of filters for shaping a character pattern in a binary image generated by the binarization circuit with different filter characteristics.
A filter selection unit for switching the selection of the plurality of filters is provided each time the threshold value is changed, and the re-recognition processing is performed by performing the filter processing with the selected filter each time the threshold value is changed. . According to the first to eleventh aspects of the present invention, since the OCR is configured as described above, when a character cannot be recognized by the recognition circuit for the first time, binarization with a changed threshold value and filter processing are performed again. The character is recognized by the re-recognition processing by the recognition circuit. Here, even if the threshold value is changed and the binary image suitable for character recognition cannot be obtained, a filter having a different filter characteristic is selected. Recognition becomes possible. Therefore, the above problem can be solved.

[0010]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS First Embodiment FIG. 1 is a block diagram of an OCR showing a first embodiment of the present invention. This OCR optically acquires the image of the form,
The image processing apparatus includes a photoelectric conversion unit 11 that generates multivalued data representing the obtained image by multivalued gradation by photoelectric conversion, and a multivalued data buffer 12 that stores the multivalued data as image data of a form. The output side of the multi-level data buffer 12 converts the image data into a binary image by binarization.
The value conversion circuit 13 is connected. These photoelectric conversion units 1
1, the multi-level data buffer 12 and the binarization circuit 13 are the same as those of the related art. An output side of the binarization circuit 13 is connected to a filter selection unit 14 which has not been provided in the related art.
Is arranged. The filter unit 15 includes a plurality of n filters 15-1, 15-2,.
15-n. These filters 15-1, 1
5-2,..., 15-n are selected by the filter selection unit 14, and the selected filter performs a filtering process on the binary data. A recognition unit 16 is connected to an output side of the filter unit 15. Recognition unit 16
Has a binary data buffer 16-1 for storing the binary image shaped by the filter processing in the same manner as in the related art, and a recognition circuit 16-2 in the same manner as in the related art. The OCR is provided with a control unit 17 for performing recognition control. Under the control of the control unit 17, the photoelectric conversion unit 11, the multi-value data buffer 12, the binarization circuit 13, the filter selection unit 1
4. The recognition unit 16 operates.

Next, the operation of each part of the OCR shown in FIG. 1 will be described. The photoelectric conversion unit 11 converts the optically input form image into digital data of multi-level gradation, and supplies the digital data to the multi-level data buffer 12. The multilevel data buffer 12 stores digital data of multilevel gradation and stores an image of a form. The binarization circuit 13 binarizes the digital data read from the multi-level data buffer 12 with a set threshold value, and generates binary data to be a binary image of the form. The filter selection unit 14 includes a filter unit 15
Obtains information on the number of scans, which is the number of times of recognition of the target character, before performing the filtering process, and selects one of the filters 15-1, 15-2,..., 15-n in the filter unit 15. . The filter unit 15 performs a filtering process on the binary data using the filter selected by the filter selecting unit 14, shapes the binary image and the character pattern therein, and provides the binary image to the binary data buffer 16-1. The binary data buffer 16-1 stores the filtered binary image. The recognition circuit 16-2 performs character recognition on the binary image stored in the binary data buffer 16-1.

The above is the basic operation of each unit.
7 transmits the binary data from the binary data buffer 16-1 to the recognition circuit 16-2 so that highly accurate recognition can be performed using the recognition result and information on the shape of the character as parameters. Perform control. FIG. 7 is a flowchart showing the recognition control of FIG. 1, and shows processes S11 to S18 of each unit according to the recognition control. Processing S11
, The photoelectric conversion unit 11 using a sensor system (not shown)
The character on the form is converted into digital data of multi-value gradation by the operation of and stored in the multi-value data buffer 12 as image data. In the binarization process of the process S12, the multi-value gradation image data is sliced by the binarization circuit 13 at a fixed threshold and converted into a binary image. In the filter selection process of the process S13, the filter selection unit 14 reads information on the number of times of scanning when generating a binary image, and selects a filter in the filter unit 15 according to the information.

FIG. 8 is an explanatory diagram of the selection of the filter shown in FIG. For example, a filter table 16T for selecting a filter is prepared in the recognition circuit 16 using the number of scans for the character to be recognized as a parameter, and a filter to be used for the scan is described for each number of scans. As a result, the filter selection unit 14 sets the filter table 16T
For example, as shown in FIG. 8, when the number of scans is the first time, the filter 15-1 and when the number of scans is two, the filter 15-
At the second and third times, 15-3 is selected. In the processing 14, the filter unit 15 uses the filter selected by the filter selection unit 14 to output the 2
The value image is filtered to form a pattern suitable for character recognition and stored in the binary data buffer 16-1. FIG. 9 is an explanatory diagram of the concept of the filtering process, and FIG.
It is a figure showing the calculation method of filter processing. FIG. 11 is a diagram illustrating an application example of the filter processing.

Each filter in the filter section 15 has, for example, 3 × 3 grid data and sets the pixel value of the pixel of interest. FIG. 9 shows three types of grid data 21, 22, and 23. The centers of the 3 × 3 grid data 21, 22, 23 are weighted with respect to the periphery, and a plurality of filters 15-1 to 15-n having different weight ratios and black point determination thresholds are prepared. This makes it possible to control the character pattern in a binary image composed of binary data to be thicker or thinner as a whole. The filter selected based on the scan count value scans the pattern of the binary data with reference to the center of the grid. The pattern A in FIG.
Although the character pattern of the broken number “2” is included, the character portion is set to 1 (for the sake of simplicity, white background 0 is omitted), and a white background (value is 0) constituting the cut portion ) As points X, Y, and Z like pattern B. Now, it is assumed that a filtering process is performed on the point X using the grid data 21. Point X is a white point and has a value of 0, and eight surrounding points are four black points and five black points. Therefore, the matrix in the pattern A centered on the point X is a matrix 24 including the data p _{11 to} p _{33 in} FIG. 10 where the black point is 1 and the white point is 0. When setting the value of the point X, each numerical value m _11- of the grid data 21 is set according to the following equation (1).
multiplying each m ₃₃ to data p ₁₁ ~p ₃₃ of the corresponding matrix 24, obtaining the sum of 9 grid of data.

[0015]

(Equation 1) The value of the point X is 5 in total. The sum of the data satisfies the condition with respect to the judgment threshold value stored in the table in advance (5 or more satisfies the condition of the black point), so the reference point is set as the black point. Similarly, the sum of the points Y and Z is 6 and 5, respectively, and is determined as a black point. A character pattern obtained by performing the above processing on all points in the cutout frame range is a pattern C in FIG.

The pattern after the filter processing is sent to the recognition circuit 16-2 via the binary data buffer 16-1, and in the recognition process of step S15, the recognition circuit 16-2 has an algorithm for determining each character category. The character recognition processing is performed using the recognition dictionary. When the recognition result becomes unreadable (when a character code is not obtained), the binarization threshold value TH is changed in step S17, and steps S12 to S16 are repeated again. When a character code is obtained by the recognition circuit 16-2, the character code is output and displayed on a display or the like as a final recognition result. As described above, in the OCR according to the first embodiment, the filter unit 15 including the plurality of filters 15-1 to 15-n having different filter characteristics and the filter selection unit 14 for selecting the filters are provided. Since the selection of the filters 15-1 to 15-n is switched based on the number of scans, even if the line width of a character does not change even if the binarization threshold TH is changed, the line width is changed by the filter characteristic. Can be changed. Therefore, the accuracy of character recognition is improved.

Second Embodiment FIG. 12 is a block diagram of an OCR showing a second embodiment of the present invention. This OCR is different from the OCR of the first embodiment in that the filter selection unit 14 is changed to the filter selection unit 31 and the recognition circuit 16-2 is changed to the recognition circuit 32. It is the same as 1. The filter selection unit 31 obtains not the number of scans but the information of the binarization threshold value TH to be changed by the binarization circuit 13,
The filter 15 in the filter unit 15 according to the threshold value TH
-1, 15-2,..., 15-n. FIG. 13 is an explanatory diagram of the selection of the filter in FIG.

In the recognition circuit 32, a filter table 32T for selecting a filter using the binarization threshold TH as a parameter
Is prepared, and a filter to be used for each binarization threshold TH is described. As a result, the filter selection unit 31 performs, for example, as shown in FIG.
If the threshold value TH is 1, the filter 15-1 is selected, if it is 2, the filter 15-2, and if it is 3, the filter 15-3 is selected. This O
In CR, character recognition is performed by the recognition control shown in FIG.
The selection of the filter to be used is changed according to the binarization threshold TH, and the recognition circuit 32 performs recognition processing on the character pattern shaped by this filter. As described above, in the second embodiment, the filter unit 1 including the plurality of filters 15-1 to 15-n having different filter characteristics is used.
5 and a filter selection unit 31 for selecting the filters 15-1 to 15-n are provided to switch the selection of the filters 15-1 to 15-n based on the binarization threshold TH. Even if the line width of a character does not change even when the threshold value TH is changed, the line width can be changed by changing the filter characteristics. Therefore, the accuracy of character recognition is improved.

Third Embodiment FIG. 14 is a block diagram of an OCR showing a third embodiment of the present invention. This OCR is different from the OCR of the first embodiment in that the filter selection unit 14 is changed to the filter selection unit 33 and the recognition circuit 16-2 is changed to the recognition circuit 34. It is the same as 1. The filter selection unit 33 detects information on the complexity of the character pattern from the binarized data instead of the number of scans, determines whether the complexity is high, normal, or low. Filters 15-1, 15-2, ..., 15-n
Is selected. The complexity can be estimated by, for example, the ratio of the number of black points to the number of white points of the binarized pattern composed of the binarized data, or the ratio of the number of black points to the size of the character pattern. FIG. 15 is an explanatory diagram of the selection of the filter in FIG.

In the recognition circuit 34, a filter table 34T for selecting a filter using the complexity as a parameter is prepared, and a filter to be used for each of high, normal, and low complexity is described. First, the filter selecting unit 33 selects, for example, the filter 15-1 to execute the filtering process, and calculates and determines the complexity from the pattern stored in the binary data buffer 16-2 based on the selection result. Then, based on the filter table 34T, the filter selecting unit 33 performs the processing shown in FIG.
5, if the complexity is high, for example, the filter 15-
1. Select the filter 15-2 if normal and 15-3 if low. Also in this OCR, character recognition is performed by the recognition control of FIG. 7, but the selection of a filter is changed in accordance with the complexity, and the recognition processing of the character pattern shaped by this filter is performed by the recognition circuit 34. As described above, in the third embodiment, the filter unit 1 including the plurality of filters 15-1 to 15-n having different filter characteristics is used.
5 and a filter selection unit 33 for selecting the filter, and the selection of the filters 15-1 to 15-n is switched based on the complexity. Therefore, even if the binarization threshold TH is changed, the character line Even when the width does not change, the line width can be changed by changing the filter characteristics. Therefore, the accuracy of character recognition is improved.

Fourth Embodiment FIG. 16 is a block diagram of an OCR showing a fourth embodiment of the present invention. This OCR is different from the OCR of the first embodiment in that the filter selection unit 14 is changed to the filter selection unit 35 and the recognition circuit 16-2 is changed to the recognition circuit 36. It is the same as 1. The filter selecting unit 35 detects information on the average line width W of the character pattern from the binarized data, instead of the number of scans, and determines whether the character pattern is thick, normal, or thin. , 15-2, ..., 15-
n is selected. The average line width W can be obtained from a binary image composed of binary data as in the conventional case. FIG. 17 is an explanatory diagram of the selection of the filter in FIG.

In the recognition circuit 36, a filter table 36T for selecting a filter using the complexity as a parameter is prepared, and a filter to be used is described for each of the large, normal, and thin average line widths W. First, the filter selection unit 35 selects, for example, the filter 15-1 to execute the filter processing, and determines the calculation and determination of the average line width W from the pattern stored in the binary data buffer 16-2 based on the selection result. Do. Based on this determination result and the filter table 36T, if the average line width W is large as shown in FIG.
1. Select the filter 15-2 if normal and 15-3 if thin. Also in this OCR, character recognition is performed by the recognition control of FIG. 7, but the selection of a filter is changed in accordance with the average line width W, and the recognition processing of the character pattern shaped by this filter is performed by the recognition circuit 36. . As described above, in the fourth embodiment, the filter unit 1 including the plurality of filters 15-1 to 15-n having different filter characteristics is used.
5 and a filter selection unit 35 for selecting the filter, and the selection of the filters 15-1 to 15-n is switched based on the average line width W. Therefore, even if the binarization threshold TH is changed, Even if the line width does not change, the line width can be changed by changing the filter characteristics. Therefore, the accuracy of character recognition is improved.

Fifth Embodiment FIG. 18 is a block diagram of an OCR showing a fifth embodiment of the present invention. This OCR differs from the OCR of the first embodiment in that the filter selection unit 14 is changed to the filter selection unit 37 and the recognition circuit 16-2 is changed to the recognition circuit 38. It is the same as 1. The filter selector 37 filters the filters 15-1 and 15-1 in the filter 15 according to both the number of scans and the binarization threshold TH.
5-2,..., 15-n. FIG. 19 is an explanatory diagram of the selection of the filter in FIG.

In the recognition circuit 38, a filter table 38T for selecting a filter using the number of scans and the binarization threshold TH as parameters is prepared, and a filter corresponding to the number of scans and the binarization threshold TH is described. deep. For example, as shown in FIG. 19, the filter selection unit 14 determines whether or not the number of scans, the binarization threshold value TH, and the filter table 38T.
Filter 15-1 when the threshold value TH is 1 and the number of scans is the first time, and Filter 1 when the number of scans is the second time
5-2, filter 15-3 when the number of scans is the third time
Select Similarly, when the binarization threshold TH is 2 or 3, the filters 15-4 to 15-9 are also set according to the number of scans.
Select Also in this OCR, character recognition is performed by the recognition control of FIG. 7, but the selection of a filter is changed according to the number of scans and the binarization threshold value TH, and the recognition circuit 38 applies the character pattern shaped by this filter. Is recognized. As described above, in the fifth embodiment, a plurality of filters 15-1 to 15-1 having different filter characteristics are provided.
5-n and a filter selection unit 37 for selecting the filter are provided, and the selection of the filters 15-1 to 15-n is switched based on the number of scans and the binarization threshold TH. Therefore, even if the line width of a character does not change even if the binarization threshold TH is changed, the line width can be changed by changing the filter characteristics. Therefore, the accuracy of character recognition is improved.

Sixth Embodiment FIG. 20 is a block diagram of an OCR showing a sixth embodiment of the present invention. This OCR is different from the OCR of the first embodiment in that the filter selection unit 14 is changed to the filter selection unit 39 and the recognition circuit 16-2 is changed to the recognition circuit 40. It is the same as 1. The filter selection unit 39 selects the filters 15-1, 15-2, 15-2, and 15-3 in the filter unit 15 according to both the complexity and the average line width W.
, 15-n are selected. FIG. 21 is an explanatory diagram of the selection of the filter in FIG.

In the recognizing circuit 40, a filter table 40T for selecting a filter is prepared by using parameters of the judgment value of the complexity and the average line width W, and a filter corresponding to the complexity and the average line width W is described. deep. The filter selecting unit 14 determines the determination values of the complexity and the average line width W and the filter table 40T.
For example, as shown in FIG. 21, the filter 15-1 is used when the average line width W is large and the complexity is high, the filter 15-2 is used when the complexity is normal, and the filter 15-2 is used when the complexity is low. Select 15-3. Similarly, when the average line width W is normal or thin, the filters 15-4 to 15-4
Select 15-9. Also in this OCR, character recognition is performed by the recognition control of FIG. 7, but the selection of a filter is changed according to the average line width W, and the recognition processing of the recognition circuit 40 is performed on the character pattern shaped by this filter. Will be As described above, in the fifth embodiment, the filter unit 15 including the plurality of filters 15-1 to 15-n having different filter characteristics and the filter selecting unit 39 for selecting the filters are provided. And the average line width W, the selection of the filters 15-1 to 15-n is switched. Therefore, even if the character line width does not change even if the binarization threshold TH is changed, the filter characteristics change. You can change the line width. Therefore, the accuracy of character recognition is improved.

Seventh Embodiment FIG. 22 is a block diagram of an OCR showing a seventh embodiment of the present invention. This OCR differs from the OCR of the first embodiment in that the filter selection unit 14 is changed to the filter selection unit 41 and the recognition circuit 16-2 is changed to the recognition circuit 42. It is the same as 1. The filter selection unit 41 filters the filters 15-1 and 15-1 in the filter unit 15 according to both the number of scans and the complexity determination value.
5-2,..., 15-n. FIG. 23 is an explanatory diagram of the selection of the filter in FIG. In the recognition circuit 42, a filter table 42T for selecting a filter using the number of scans and the determination value of the complexity as parameters is prepared, and a filter corresponding to the number of scans and the complexity is described. Filter selector 14
Based on the determination values of the number of scans and the complexity, and the filter table 42T, for example, as shown in FIG. 23, the filter 15-1 is used when the complexity is high and the number of scans is first, and the filter 15 is used when the number of scans is second. -2, when the number of scans is 3, the filter 15-3 is selected. Similarly, when the complexity is normal or low, the filters 15-4 to 15-9 are selected according to the number of scans.

Also in this OCR, character recognition is performed by the recognition control shown in FIG. 7, but the selection of a filter is changed in accordance with the number of scans and the complexity, and the recognition circuit 42 A recognition process is performed. As described above, in the seventh embodiment, the filter unit 15 including the plurality of filters 15-1 to 15-n having different filter characteristics and the filter selection unit 41 for selecting the filter are provided. And the degree of complexity, the selection of the filters 15-1 to 15-n is switched. Therefore, even if the line width of the character does not change even if the binarization threshold value TH is changed, the line is changed by the change in the filter characteristics. You can change the width. Therefore, the accuracy of character recognition is improved.

Eighth Embodiment FIG. 24 is a block diagram of an OCR showing an eighth embodiment of the present invention. This OCR is different from the OCR of the first embodiment in that the filter selection unit 14 is changed to the filter selection unit 43 and the recognition circuit 16-2 is changed to the recognition circuit 44. It is the same as 1. The filter selection unit 43 determines the filter 15-in the filter unit 15 according to both the number of scans and the determination value of the average line width W.
, 15-n are selected. FIG. 25 is an explanatory diagram of the selection of the filter in FIG. In the recognition circuit 44, a filter table 44T for selecting a filter using the number of scans and the determination value of the average line width W as parameters is prepared, and a filter corresponding to the number of scans and the average line width W is described. The filter selection unit 14 determines, for example, FIG. 25 based on the determination values of the number of scans and the average line width W and the filter table 44T.
, The filter 15-1 when the average line width W is large and the number of scans is the first, the filter 15-2 when the number of scans is the second, and the filter 1 when the number of scans is the third
Select 5-3. Similarly, when the average line width W is normal or low, the filters 15-4 to 15-1 are changed according to the number of scans.
Select 5-9.

Also in this OCR, character recognition is performed by the recognition control shown in FIG. 7, but the selection of a filter is changed according to the number of scans and the average line width W, and the character pattern shaped by this filter is recognized. The recognition processing of the circuit 44 is performed. As described above, in the eighth embodiment, a plurality of filters 15-1 to 15-n having different filter characteristics are provided.
Is provided, and a filter selection unit 43 for selecting the filter is provided, and the selection of the filters 15-1 to 15-n is switched based on the number of scans and the average line width W. Even when the line width of a character does not change even when the threshold value TH is changed, the line width can be changed by changing the filter characteristics. Therefore, the accuracy of character recognition is improved.

Ninth Embodiment FIG. 26 is a block diagram of an OCR according to a ninth embodiment of the present invention. This OCR is different from the OCR of the first embodiment in that the filter selection unit 14 is changed to the filter selection unit 45 and the recognition circuit 16-2 is changed to the recognition circuit 46. It is the same as 1. The filter selection unit 45 determines the filter 15-in the filter unit 15 according to both the binarization threshold value TH and the determination value of the average line width W.
, 15-n are selected. FIG. 27 is an explanatory diagram of the selection of the filter in FIG. In the recognition circuit 46, the binarization threshold TH
A filter table 46T for selecting a filter using parameters and the determination value of the average line width W as parameters is prepared.
A filter corresponding to H and the average line width W is described.
The filter selecting unit 14 is configured, for example, as shown in FIG.
7, when the average line width W is large and the binarization threshold TH is 1, the filter 15-1 is used, and when the binarization threshold is 2, the filter 1 is used.
When 5-2 and the binarization threshold TH are 3, the filter 15-3 is selected. Similarly, when the average line width W is normal or low, the filters 15-4 to 15- according to the binarization threshold value TH.
Select 9.

Also in this OCR, character recognition is performed by the recognition control shown in FIG. 7, but the selection of a filter is changed in accordance with the binarization threshold value TH and the average line width W, and the character pattern shaped by this filter is converted to a character pattern. On the other hand, the recognition processing of the recognition circuit 46 is performed. As described above, in the ninth embodiment,
A plurality of filters 15-1 to 15- having different filter characteristics
n, and a filter selection unit 45 for selecting the filter are provided to switch the selection of the filters 15-1 to 15-n based on the binarization threshold TH and the average line width W. Therefore, even if the line width of a character does not change even if the binarization threshold TH is changed, the line width can be changed by changing the filter characteristics. Therefore, the accuracy of character recognition is improved.

Tenth Embodiment FIG. 28 is a block diagram of an OCR showing a tenth embodiment of the present invention. This OCR is different from the OCR of the first embodiment in that the filter selection unit 14 is changed to the filter selection unit 47 and the recognition circuit 16-2 is changed to the recognition circuit 48. It is the same as 1.
The filter selecting unit 47 determines the filter 15-in the filter unit 15 according to both the binarization threshold TH and the complexity determination value.
, 15-n are selected. FIG. 29 is an explanatory diagram of the selection of the filter in FIG. In the recognition circuit 48, the binarization threshold TH
A filter table 48T for selecting a filter is prepared using the parameter and the determination value of the complexity as parameters, and a filter corresponding to the binarization threshold TH and the complexity is described. Based on the binarization threshold TH and the determination value of the complexity, and the filter table 48T, the filter selection unit 14 determines whether the filter 1 has a high complexity and the binarization threshold TH is 1 as shown in FIG.
5-1: The filter 15-2 is selected when the binarization threshold is 2, and the filter 15-3 is selected when the binarization threshold TH is 3. Similarly, when the complexity is normal or low, the binarization threshold TH
Select the filters 15-4 to 15-9 according to.

Also in this OCR, character recognition is performed by the recognition control shown in FIG. 7, but the selection of a filter is changed according to the binarization threshold value TH and the complexity, and the character pattern shaped by this filter is The recognition processing of the recognition circuit 48 is performed. As described above, in the ninth embodiment, a plurality of filters 15-1 to 15-n having different filter characteristics are provided.
And a filter selection unit 47 for selecting the filter, the selection of the filters 15-1 to 15-n is switched based on the binarization threshold value TH and the complexity. Even if the line width of a character does not change even when the threshold value TH is changed, the line width can be changed by changing the filter characteristics. Therefore, the accuracy of character recognition is improved. Note that the present invention is not limited to the above embodiment, and various modifications are possible. For example, in the first to tenth embodiments, the filters arranged in the filter unit 15 are 15 ₁ to
The number of 15 _n is arbitrary, and if the number corresponding to the number of ranks of the parameters is provided, the same effect as above can be obtained.

[0035]

As described in detail above, first to first
According to the first aspect of the present invention, there is provided a photoelectric conversion unit, a multi-valued data buffer, a binarization circuit, a filter unit, and a recognition circuit. In an optical character reading apparatus for recognizing characters by re-recognition processing after performing conversion and re-filtering, the filter unit has a configuration having a plurality of filters having different filter characteristics. And a filter selection unit that switches the selection of the filter. Therefore, even if the line width of the character pattern in the binary image does not change even if the threshold value is changed, a character pattern suitable for character recognition can be obtained with the filter characteristics. Since it can be formed, highly accurate character recognition becomes possible.

[Brief description of the drawings]

FIG. 1 is a configuration diagram of an OCR according to a first embodiment of the present invention.

FIG. 2 is a configuration diagram showing a main part of a conventional OCR.

FIG. 3 is a flowchart showing the recognition control of FIG. 2;

FIG. 4 is an explanatory diagram of calculating an average line width.

FIG. 5 is an explanatory diagram of a change in a character pattern according to a binarization threshold.

FIG. 6 is an explanatory diagram of the problem in FIG. 2;

FIG. 7 is a flowchart showing the recognition control of FIG. 1;

FIG. 8 is an explanatory diagram of selection of a filter in FIG. 1;

FIG. 9 is an explanatory diagram of the concept of filter processing.

FIG. 10 is a diagram illustrating a calculation method of a filtering process.

FIG. 11 is a diagram illustrating an application example of a filtering process.

FIG. 12 is a configuration diagram of an OCR according to a second embodiment of the present invention.

13 is an explanatory diagram of selection of the filter in FIG.

FIG. 14 is a configuration diagram of an OCR according to a third embodiment of the present invention.

FIG. 15 is an explanatory diagram of the selection of the filter in FIG. 14;

FIG. 16 is a configuration diagram of an OCR according to a fourth embodiment of the present invention.

FIG. 17 is an explanatory diagram of selection of the filter in FIG. 16;

FIG. 18 is a configuration diagram of an OCR according to a fifth embodiment of the present invention.

FIG. 19 is an explanatory diagram of selection of the filter in FIG. 18;

FIG. 20 is a configuration diagram of an OCR according to a sixth embodiment of the present invention.

FIG. 21 is an explanatory diagram of selection of the filter in FIG. 20;

FIG. 22 is a configuration diagram of an OCR according to a seventh embodiment of the present invention.

FIG. 23 is an explanatory diagram of selection of the filter in FIG. 22;

FIG. 24 is a configuration diagram of an OCR according to an eighth embodiment of the present invention.

FIG. 25 is an explanatory diagram of selection of the filter in FIG. 24;

FIG. 26 is a configuration diagram of an OCR according to a ninth embodiment of the present invention.

FIG. 27 is an explanatory diagram of selection of the filter in FIG. 26;

FIG. 28 is a configuration diagram of an OCR according to a tenth embodiment of the present invention.

FIG. 29 is an explanatory diagram of the selection of the filter in FIG. 28.

[Explanation of symbols]

Reference Signs List 11 photoelectric conversion unit 12 multi-value data buffer 13 binarization circuit 14, 31, 33, 35, 37, 39, 41, 43, 4
5, 47 filter selection unit 15 filter unit 15-1 to 15-n filter 16 recognition unit 16-1 binary data buffer 16-2, 32, 34, 36, 38, 40, 42, 4
4, 46, 48 recognition circuit 17 control unit

Claims (11)

[Claims] A photoelectric conversion unit that optically captures an image of a form on which characters are written, photoelectrically converts the image, and generates multi-value data representing the image of the form in multi-valued gradation; A multi-level data buffer for storing value data; a binarization circuit for binarizing the multi-level data read from the multi-level data buffer with a set threshold to generate a binary image corresponding to the form; A filter unit that performs a filtering process on a binary image output from the binarization circuit to shape a character pattern in the binary image, and performs a recognition process on the filtered binary image, And a recognition circuit for recognizing the character. If the character cannot be recognized by the recognition circuit, re-recognition by the recognition circuit after performing the binarization with the threshold changed and the filter processing again Processing recognizes the characters In the optical character reading device, the filter unit is configured to have a plurality of filters for respectively shaping a character pattern in a binary image generated by the binarization circuit with different filter characteristics, and each time the threshold value is changed A filter selection unit for switching the selection of the plurality of filters is provided, and the re-recognition process is performed by performing the filter process with the selected filter every time the threshold value is changed. Character reader. 2. The optical character reading device according to claim 1, wherein the filter selection unit is configured to select a plurality of filters of the filter unit based on the number of times of the re-recognition processing. 3. The optical character reading device according to claim 1, wherein the filter selection unit is configured to select a plurality of filters of the filter unit based on the threshold value to be changed. 4. The optical character according to claim 1, wherein the filter selection unit is configured to select a plurality of filters of the filter unit based on the complexity of a character pattern in the binary image. Reader. 5. The optical system according to claim 1, wherein the filter selection unit is configured to select a plurality of filters of the filter unit based on an average line width of a character pattern in the binary image. Character reader. 6. The optical system according to claim 1, wherein the filter selection unit is configured to select a plurality of filters of the filter unit based on the number of times of the re-recognition processing and the threshold value to be changed. Character reader. 7. The filter selecting unit is configured to select a plurality of filters of the filter unit based on a line width of a character pattern in the binary image and a complexity of the character pattern. The optical character reading device according to claim 1. 8. The filter selecting unit is configured to select a plurality of filters of the filter unit based on the number of times of the re-recognition processing and the complexity of a character pattern in the binary image. The optical character reading device according to claim 1. 9. The filter selecting unit is configured to select a plurality of filters of the filter unit based on the number of times of the re-recognition processing and an average line width of a character pattern in the binary image. The optical character reader according to claim 1. 10. The apparatus according to claim 1, wherein the filter selecting unit selects a plurality of filters of the filter unit based on the threshold value to be changed and an average line width of a character pattern in the binary image. Item 1. The optical character reading device according to Item 1. 11. The apparatus according to claim 1, wherein the filter selecting unit selects a plurality of filters of the filter unit based on the threshold value to be changed and a complexity of a character pattern in the binary image. 2. The optical character reader according to 1.