JPS6027436B2 - Character recognition correction method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a character recognition correction method that corrects the center position of a character when character recognition is performed on a memory to read the character lines of a form.

Conventionally, a scanner of an optical character reader (hereinafter referred to as OCR) detects a character pattern for one line of a form as video data, stores it in memory, and cuts out and recognizes the character pattern on this memory. Character recognition methods are often used.

In this case, as shown in FIG. 1, when the form 2 is fed in the direction 5 to the scanning surface 1 by the optical scanner of the OCR device and the line mark 3 is detected by the sensor and set,
As shown in the figure, it may be set in a slanted position. The character line 4 to be detected on the form 2 is written so that its center line is on the line PP' passing through the corresponding line marks 3, 32 on both sides, that is, perpendicular to both edges of the form. Therefore, it forms a certain angle 8 with the scanning direction PQ, which is orthogonal to the feeding direction 5 of the form 2. In this relationship, the video data scanned by the scanner is stored in the memory, and characters are cut out and recognized in the scanning direction PQ in the same relationship as in FIG. 1 with respect to the stored character pattern PP' of the character line. Normally, the setting for scanning the form 2 is performed only with the line sensor 3 on one side, and if the stop position deviation in this detection is within a certain range, rescanning is further performed when recognition is not possible, and when determining the epistemology. The positional deviation has been corrected. However, since no correction is made for the deviation of angle 8 due to the inclination of form 2, in the case of this type of deviation, many reject characters occur, especially when character patterns near the left edge are recognized. Ru. If we examine in detail the character pattern that is ejected in this way, as shown in Figure 2, the character pattern in the case of a normal set of form 2 in the scanning direction PQ during recognition in memory is shown by a dotted line frame 8. If you are having trouble with Form 2,
There is a vertical shift of the center of the character corresponding to the angle saw and a slight rotational shift.

Therefore, when the vertical shift of the center of the character is corrected, only a slight rotational shift remains as shown by the solid line frame 7 in the figure. Naturally, a slight rotational shift also appears in the character 6, but it is sufficiently recognizable as long as the angle a is not too large. Such correction should be able to repair many rejected characters caused by this type of shift. SUMMARY OF THE INVENTION An object of the present invention is to provide a character recognition correction method that can effectively correct the irregularity of character lines in a form and effectively correct the problem in character recognition.

In order to achieve the above object, the character recognition correction method of the present invention stores a character pattern for one line of a form detected by a scanner of an optical character discussion device in a memory, and performs character recognition discussion on the memory. In a character recognition method, means for detecting a line mark at one end provided at one end of a form, means for detecting a line mark at the other end provided at the other end of the form, and a positional deviation between the line mark at one end and the line mark at the other end. a comparison detection means for detecting the positional deviation, a division means for dividing the positional deviation amount detected by the comparison detection means by a predetermined number of characters for one line to calculate the deviation amount due to the pattern for each character; At the time of character recognition processing after storing the character pattern for a line in memory and detecting all the other line marks mentioned above, the one end line mark position information, the deviation amount information per character, and the number of the character to be recognized are determined. calculation means for calculating a center position correction amount for each character based on information indicating whether the character string is slanted; In contrast, the center position correction amount is applied to each character, and recognition and reading are performed while sequentially correcting the center of each character by the amount of deviation for each character.

The present invention will be described in detail below with reference to examples.

FIG. 3 is an explanatory diagram of the configuration and functions of the main parts of the present invention.

A feature of the present invention is that the above-mentioned rejection problem caused by the inclination of the form is solved by correcting the recognition exclusively in memory, while leaving the form and scanner as they are. The memory 11 of the recognition device is shown as the diamond part. In the figure, a character pattern for one line of a form is stored in a memory 11 as video data by a scanner 10 of an OCR device.

Assume now that the character patterns Y12, -1 are set as shown in the figure because of the inclination of the form. That is, mark 1 corresponds to line mark 3 at the right end of form 2 in FIG.
3, and scanning is performed from this position, character pattern video data 12n to 12, are stored at an angle of 8, and a mark 132 corresponding to the leftmost line mark 32 is detected. Let the positions of these marks 131 and 132 be represented by a (bit) and b (bit), respectively, and divide the bit difference d corresponding to this inaccuracy by the number of characters n to calculate the deviation △ due to the slope for each character. . That is, △ = bit) 3a (bit) one trick {・} is obtained, and the character center is sequentially moved upward or downward by △ for each letter, and the recognition discussion of the character pattern is performed.

In this way, as explained in FIG. 2, a character pattern with a slight rotational deviation compared to the normal case can be recognized, and recognition is possible within a range where the angle a is not too large.

FIG. 4 is an explanatory diagram of an embodiment of the present invention comprising a recognition device including the memory 11 of FIG.

In the figure, a line mark sensor 21 detects the positional information (address) of point a of the line mark on the right side of the form and stores it in a register 22.

This point becomes the center position of the first character pattern in the scan. This positional information of point a is input to the comparator 24, and is branched and sent to an adder 31, which will be described later. Further, as shown in FIG. 3, video data of a character pattern for one line of a form is stored in the memory 1 of the recognition device 29 by the scanner 10. Then, scan the leftmost line mark b of the form.
When the point is reached, the position information (address) of point b is detected by the line mark sensor 23 and input to the comparator 24 to obtain (ba) in the above equation {1}. This output (b-a
) is input into the divider 27 and divided by the number n of characters retrieved from the format register 25 and held in the memory 26 to find the center shift amount for each character. On the other hand, at the same time as point b is detected by the line mark sensor 23, a start signal is sent to the recognition device 29, character re-recognition is started character by character on the memory 11, and the center position is moved character by character.

Therefore, a character counter 30 is provided that counts each character after each character is recognized, and this count value and the output of the divider 27 (△=sheep) are provided.
By multiplying by the multiplier 28, the amount of deviation from the center point a of each character pattern is calculated.

Therefore, the output of this multiplier 28 is input to the adder 31,
By adding this to the position information of point a described above, a correction value for the center position of each character can be obtained. If this is held in the register 32 and supplied as a center for each character recognition by the recognition device 29, most of the characters can be recognized with only a slight rotational deviation.
As explained above, according to the present invention, by correcting the center position of each character when character recognition is performed in memory in order to reduce the trouble of character lines on a form, In this case, most of the characters can be recognized, and most of the characters can be repaired, whereas in the past, this type of discrepancy often resulted in overlapping characters.

The reason for this is that if the inclination of the form is too large, it will not be considered normal paper feeding and will be removed separately. In this way, it greatly contributes to reducing the number of rejected characters in character recognition, and the efficiency of character recognition can be improved.

[Brief Description of the Drawings] Fig. 1 is an explanatory diagram of the problem of document scanning in an OCR device, Fig. 2 is a schematic explanatory diagram of the principle of the present invention, and Fig. 3 is an illustration of the configuration and functions of the main parts of the present invention. Explanatory diagram, FIG. 4 is an explanatory diagram showing the configuration of an embodiment of the present invention, in which 2 is a form, 3, 3
2 is a line mark, 4 is a character line, 10 is a scanner, 11 is a memory, 122 to 12n are character patterns, 131, 13
2 is a detection mark, 21 and 23 are line mark sensors, 2
2 and 32 are registers, 24 is a comparator, 25 is a format register, 26 is a memory, 27 is a divider, 28 is a multiplier, 29 is a recognition device, 30 is a character number counter, and 31 is an adder. Figure 1 Figure 2 Figure M Ship diagram and dimensions

Claims (1)

[Claims] 1 In a character recognition method in which a character pattern for one line of a form detected by a scanner of an optical character reading device is stored in a memory and characters are recognized and read on the memory, one line mark provided at one end of the form is means for detecting, and means for detecting the other end line mark provided at the other end of the form;
A comparison detection means for detecting a positional deviation between the one end line mark and the other end line mark, and a slope for each character obtained by dividing the amount of positional deviation detected by the comparison detection means by a predetermined number of characters for one line. and a calculation means for calculating the amount of deviation due to the one-line mark position information and the one-line mark position information during character recognition processing after storing the character pattern for one line in a memory and detecting the other-end line mark. calculation means for calculating the center position correction amount of the recognition target character for each character based on the shift amount information and the information indicating the number of the recognition target character, In this case, the center position correction amount is given to each character to the character information stored in the memory, and the recognition reading is performed while sequentially correcting the center of each character by the amount of deviation for each character. Character recognition correction method.