JPH07160808A - Frame cutting method in character recognition - Google Patents

Abstract

(57) [Abstract] [Purpose] Paper that the position of the frame in which the character is written is not accurately specified by searching for the position of the frame in which the character is written, and determining and correcting the tilt of the frame. Recognize the characters written in with a high recognition rate. [Structure] Image information is read from a sheet 1 in which characters, numbers, symbols, etc. are written in a frame 2, and the frame 2 is searched from a set position below (or above) the image information as shown by an arrow in the figure. When the frame 2 is found, the frame 2 is traced and the four corners of the frame 2 are found. 4 corners obtained,
Then, the position and the inclination of the frame 2 are obtained from, and the inclination of the frame 2 is rotationally corrected. Then, from the rotation-corrected frame, letters, numbers,
Pattern 3 such as a symbol is cut out and character recognition is performed.

Description

Detailed Description of the Invention

[0001]

[Industrial application] Characters, numbers, symbols, etc. written on the frame of OCR paper etc. (Hereinafter, these are collectively referred to as "characters".
Character recognition processing is performed in various fields in which character recognition is performed by reading a character pattern from the read image information and extracting a character pattern from the read image information.

In the above character recognition processing, the character pattern cannot be cut out unless the position of the frame for writing the character is specified, and if the read character is tilted, the character cannot be cut out accurately and the character recognition rate. Is reduced. INDUSTRIAL APPLICABILITY The present invention provides a frame cutting process for finding a position of a frame and cutting out a character pattern in a character recognition process in which a character written on a sheet in which the position of the frame for writing the character is not specified is read and recognized In particular, the present invention finds four points at the corners of a character frame,
The present invention relates to a frame cutting processing method in character recognition in which a character is cut out in a correct direction by correcting the inclination of the character frame and a high character recognition rate can be obtained.

[0003]

2. Description of the Related Art OCR paper is usually provided with a character entry frame for entering characters, and a position information mark for identifying the position of the character entry frame is recorded on the periphery of the paper. . Then, the character is written in the predetermined character entry frame. FIG. 13 is a diagram showing an example of the above-mentioned OCR paper, and as shown in FIG.
A frame 101 for writing characters and a position information mark 102 are recorded on the.

When reading the written characters, the position of the frame 101 is calculated based on the position information mark 102, the characters in the frame 101 are cut out, and the character recognition processing is performed. In the conventional method as described above, the OCR
Characters cannot be cut out without using paper such as paper that has a specified writing position. It is not possible to cut out and recognize characters written on ordinary paper that does not have a frame at a specific position. It was difficult.

In particular, when reading a normal sheet without a position information mark with a scanner or the like, the inclination of the sheet cannot be determined. Therefore, when the sheet is read obliquely and the frame is inclined, Even if the position of the frame in which the character is written is specified on the paper, the character cannot be cut out in the correct direction, and the recognition rate of the character decreases.

[0006]

SUMMARY OF THE INVENTION The present invention has been made to solve the above-mentioned problems of the prior art.
A first object of the present invention is to provide a frame cutting processing method in character recognition which can recognize a character written on a sheet in which the position of the frame for writing the character is not specified with a high recognition rate.

A second object of the present invention is to search for the position of the frame of the paper, find the inclination of the frame and correct it, so that even if the frame is read with an inclination, the characters are read in the correct direction. An object of the present invention is to provide a frame cutting processing method in character recognition that can be cut out.

[0008]

FIG. 1 is a principle view of the present invention. In FIG. 1, 1 is a sheet having a frame for writing characters, numbers, symbols, etc., 2 is a frame recorded on the sheet 1, ,
,, are the four corners of the frame 2 obtained by searching, and 3 is a pattern of characters, numbers, symbols, etc. cut out from the image information.

In order to solve the above problems, as shown in FIG. 1, the invention of claim 1 of the present invention comprises a frame 2 for writing characters, numbers, symbols, etc., and the frame 2 is provided with characters, numbers, symbols, etc. In the frame cutting processing method in character recognition, which reads the image information from the sheet 1 in which is written, cuts out the pattern 3 of characters, numbers, symbols, etc. from the read image information to perform character recognition, and searches for the image information. Find a part of the frame 2, trace the frame 2 you found, 4 corners of the frame,
,,, are found, the position of the frame 2 on the paper is obtained based on the information of the four corners of the frame ,,,, and the pattern 3 such as the characters, numbers, and symbols in the obtained frame 2 is cut out. .

According to a second aspect of the present invention, in the first aspect of the invention, the inclination of the frame 2 is obtained from the information of the found four corners of the frame, and the frame 2 is rotationally corrected based on the obtained inclination. A pattern 3 of characters, numbers, symbols, etc. is cut out from the frame 2 which has been subjected to the rotation correction.

[0011]

In FIG. 1, the image information is read from the sheet 1 in which characters, numbers, symbols, etc. are written in the frame 2, and the frame 2 is read from the setting position below (or above) the image information as shown by the arrow in the figure. To explore. When the frame 2 is found, the frame 2 is traced to find the four corners of the frame 2 ,.

The position and the inclination of the frame 2 are obtained from the obtained four corners, and the inclination of the frame 2 is rotationally corrected. Then,
Characters are recognized by cutting out patterns such as letters, numbers, and symbols from the rotation-corrected frame. In the invention of claim 1 of the present invention, as described above, the image information is searched and the frame 2 is searched.
Find a part of the frame, trace the frame 2 you found, and then add the four corners of the frame
Find out ,,, and the four corners of the frame ,,,
Since the position of the frame 2 on the paper is obtained based on the information of, the character pattern can be accurately cut out even if the character is written on the paper in which the position of the frame for writing the character is not specified. Characters can be recognized with a high recognition rate.

According to the second aspect of the present invention, the four corners of the frame found in the first aspect of the invention are ...
Since the inclination of the frame 2 is obtained from the information of the frame 2 and the rotation of the frame 2 is corrected based on the obtained inclination, the character can be cut out in the correct direction even when the frame is read with the inclination. The recognition rate can be improved.

[0014]

FIG. 2 is a diagram showing the overall configuration of a system according to an embodiment of the present invention. In FIG. 2, 11 is a scanner and 11 is a scanner.
a is an optical sensor such as a CCD that optically reads image information recorded on a sheet, 12 is a quantizer that quantizes the image information read by the optical sensor 11a, and 13 is quantized information as image image data. An image memory 14 to be stored is a frame search unit described later, and the frame search unit 14 searches for the position of the frame in the read image information and obtains its inclination.

A tilt correction unit 15 corrects the frame tilt based on the frame tilt obtained by the frame search unit 14. 16
Is a cutout processing unit for cutting out a character pattern portion from the frame, and 17 is a character recognition unit for performing character recognition on each cut-out character pattern. In the figure, the optical sensor 11a of the scanner 11 reads characters written in a frame provided at an arbitrary position on the paper as image information. The read image information is converted into a digital signal by the quantizer 12 and stored in the image memory 13 as image image data.

The frame search unit 14 uses a sensor described later to
The position of the frame in the image data stored in the image memory is searched. And when you discover a part of the frame,
Trace the frame, find the four corners of the frame, and find the tilt of the frame. The tilt correction unit 15 rotationally corrects the tilt of the frame based on the tilt of the frame obtained by the frame search unit 14. FIG. 3 is a diagram showing the concept of frame search and tilt correction processing in the frame search unit 14 and the tilt correction unit 15 described above. The frame search unit 14 starts from the frame search start point A as shown in FIG. When the frame search is started as shown by the arrow in the figure and a part of the frame is found at in the figure, the frame is traced as shown in and the first corner of the frame is found at.

Similarly, the frame is changed to →→→→
Follow the frame in the order of → to find the four corners. Then, the position of the character frame and the inclination of the frame are obtained from the obtained corner position, and the inclination is rotationally corrected as shown in FIG. The cutout processing unit 16 cuts out a character pattern from the frame whose tilt is corrected by the tilt correction unit 15, and further cuts out a character pattern for each character from the cut out character pattern. The character recognition unit 17 performs character recognition on each character pattern cut out by the cutout processing unit 16.

FIGS. 4 and 5 are flowcharts showing the character recognition processing in this embodiment. The character recognition processing in this embodiment will be described with reference to FIGS. In steps S1 and S2 of FIG. 4, when the same data as the frame in the image data is searched and a part of the frame is found, in step S3, the frame line is changed depending on the inclination of the frame (referred to as a tracing angle). Estimate a certain place and follow the frame while averaging the tracing angles.

If the same data as the frame in the traced direction disappears in step S4, it is assumed that the first corner is found, and in step S5, the direction is changed to the vertical direction and the frame is traced again. If the same data as the frame disappears in the traced direction in step S6, it is assumed that the second corner is found, and in step S7, the same as above.
The direction is changed to the vertical direction and the frame is traced again.

In step S8 of FIG. 5, if the same data as the frame in the traced direction disappears, the third corner is found, and in step S9, the direction is changed to the vertical direction and the frame is traced again. To go. If the same data as the frame in the traced direction disappears in step S10, it is determined in step S11 whether the obtained four points satisfy the conditions (side length, aspect ratio) of the rectangle as the frame. If not satisfied, the process returns to step S1 and the above process is repeated.

If the conditions are met, step S
In step 12, the affine transformation is performed in order to find the tilt of the frame and correct the rotation of the frame. Next, in step S13, a character is cut out based on the four corners of the rotation-corrected frame, and character recognition processing is performed on the cut-out character pattern in step S14.

FIG. 6 is a diagram showing a sensor for searching a frame and a state of the frame search by the sensor, FIGS. 7 to 8 are diagrams showing a process (1) for finding a part of the frame, and FIGS. 9 to 12 are the frames. FIG. 7 is a diagram showing a process (2) of tracing a frame and a process (3) of finding a corner of a frame, and FIGS. 7, 8, and 10 to 12 are diagrams in which image image data is enlarged to a dot level. Next, an example of the frame search processing in the frame search unit 14 will be described with reference to FIGS. (1) Processing for discovering part of frame FIG. 6A is a diagram showing "a sensor for discovering a part of a frame", and each sensor bit in the sensor corresponds to one dot of image image data. Therefore, the width of the sensor bit in this embodiment is 1 byte width as shown in FIG.

In order to search the frame by the above sensor, the dot of the image image data at the sensor bit position is determined. At that time, the sensor bit responds when the data of the dot is the same data as the frame. For example, when the frame is black, the sensor bits respond when the dots of the image data at the sensor bit positions are the same black.

The width of the sensor bit is related to the processing speed, the ability to discriminate from dust, etc. The width of the sensor bit being 1 byte wide contributes to the processing speed, and small dust is mistaken as a frame. Although it is suitable to make it difficult to do so, the width of the sensor bit can be changed depending on the size of the frame in the image data. FIG. 6B is a diagram showing a frame finding process by the sensor. As shown in FIG. 6B, the sensor is positioned below (or above) the image data.
The frame is searched by moving one step from the setting position of.

Then, when the left or right sensor bit has the same data as the frame, the reacted sensor bit is moved in the traveling direction until the same data as the frame disappears, and the number of bits between them is calculated, Comparison is made with a predetermined line width condition range [see FIG. 7 (a)]. That is, it is confirmed whether the line width of the frame is within the tolerance. Next, the sensor bit that is not responding (the bit on the right side in FIG. 7A) is moved by a set amount in the traveling direction from the height at which the responding sensor bit responded, and a search is performed. If there is a response to, the number of dots is calculated and compared with the line width condition range as described above (see FIG. 7B).

Further, the number of dots on the left and right (line width)
Is compared with a predetermined left / right line width error condition range. If the calculated number of dots clears the line width condition range, and the difference clears the left and right line width error condition range, the frame tilt is predicted by the reaction order of the left and right sensor bits,
The inclination degree (referred to as a tracing angle) is obtained from the dot difference between the reaction points of the left and right sensor bits [see FIG. 8]. Also, the average line width of the left and right line widths is obtained.

Then, since the data to which the sensor bit has reacted is the possibility of being the data in the frame, it starts tracing the data. By performing the above processing, it is difficult to react to dust (noise), a part of the frame can be found at a relatively high processing speed, and the characteristics of the frame can be predicted. (2) Processing to trace the frame After discovering a part of the frame, the direction to trace is determined by the inclination of the frame. FIG. 9 is a diagram showing the relationship between the inclination of the frame and the tracing direction. In the figure, a circle indicates that a part of the frame is found,
● indicates that the corner of the frame was found.

As shown in FIG. 9, when the inclination of the frame is downward leftward, the frame is traced counterclockwise, and when it is downward rightward, the frame is traced clockwise.
That is, the direction to follow is determined in order to reduce the influence of dust on the outside of the frame. Next, the size of the "sensor for tracing the frame" is determined in consideration of the fact that the tracing angle is a value for the size of the sensor used in (1) above. The size of the "sensor for tracing the frame" is preferably the same as that of the "sensor for finding a part of the frame".

FIG. 10 is a diagram showing the relationship between the sensor and the image image data. As shown in FIG. 10, the sensor bit of the "sensor for tracing the frame" is provided at the end in the tracing direction, and the sensor -The bit reacts when the dot of the image data is the same data as the frame, like the sensor bit of "Sensor for finding part of frame".

When tracing the frame, as shown in FIG. 11A, the sensor estimates the location of the frame line based on the tracing direction and the tracing angle, and searches inside and outside of the frame by a set amount. When the sensor bit reacts, the line width is calculated as in the case of (1) above, and if it is within the tolerance of the average line width obtained in (1) [see FIG. 11 (b)], It is recognized as a part and the line width is fed back to the average line width. Further, the dot difference between the point at which the current sensor has reacted and the reaction point immediately before it is fed back to the tracing angle, and the process proceeds to the next search. (3) Processing to find the corner of the frame When the sensor moves by tracing the frame line and there is no reaction to the estimated search range, it is considered that the corner has entered the existing range, and the sensor bit of the current sensor position Move the sensor bit one bit at a time from the opposite side,
Based on the condition of the line width, the frame line is traced in the same manner as above. Then, the position where there is no reaction is defined as a corner [see FIG. 12].

As described above, the angle is found, and each time the angle is found, the direction of the search is changed to a right angle, and the processing shown in (2) and (3) is performed while feeding back the average line width and the tracing angle. Do 3 more times. When the four corners are found, the vertical length and the horizontal length of the frame are calculated, and the size, ratio, etc. are compared with preset conditions, and if satisfied, the frame is recognized.

By performing the above-mentioned processes (2) and (3), it is possible to detect the tracing angle of the frame and to recognize the frame stably without being relatively affected by the dust on the outside. Become.

[0033]

As described above, according to the present invention, the image information is searched to find a part of the frame, and the four corners of the frame are found by tracing the found frame to determine the position of the frame on the sheet. Therefore, even if the character is written on the paper in which the position of the frame for writing the character is not specified, the character pattern can be accurately cut out, and the character can be recognized with a high recognition rate. .

Further, by obtaining the tilt of the frame from the information of the four corners of the found frame and correcting the rotation of the frame based on the found tilt, the character is correct even if the frame is read with tilting. It can be cut out in the direction, and the character recognition rate can be improved.

[Brief description of drawings]

FIG. 1 is a principle diagram of the present invention.

FIG. 2 is a diagram showing an overall configuration of a system according to an embodiment of the present invention.

FIG. 3 is a diagram illustrating a concept of frame search and inclination correction processing according to an embodiment.

FIG. 4 is a flowchart of an embodiment of the present invention.

FIG. 5 is a flowchart (continuation) of the embodiment of the present invention.

FIG. 6 is a diagram showing a sensor for searching a frame and a state of frame search by the sensor.

FIG. 7 is a diagram showing a process of finding a part of a frame.

FIG. 8 is a diagram (continuation) showing a process of discovering a part of a frame.

FIG. 9 is a diagram showing a process of tracing a frame.

FIG. 10 is a diagram (continuation) showing a process of tracing a frame.

FIG. 11 is a diagram (continuation) showing a process of tracing a frame.

FIG. 12 is a diagram illustrating a process of finding a corner of a frame.

FIG. 13 is a diagram showing an example of conventional OCR paper.

[Explanation of symbols]

1 Paper 2 Frame 3 Patterns of Characters, Numbers, Symbols Cut Out 11 Scanner 11a Optical Sensor 12 Quantization Section 13 Image Memory 14 Frame Search Section 15 Tilt Correction Section 16 Cutout Processing Section 17 Character Recognition Section

Claims (2)

[Claims] 1. A sheet provided with a frame (2) for writing characters, numbers, symbols, etc., in which characters, numbers, symbols, etc. are written.
In the frame cutting processing method for character recognition, in which the image information is read from (1), and the pattern (3) of characters, numbers, symbols, etc. is cut out from the read image information and character recognition is performed, the above image information is searched to create a frame. Find a part of (2), trace the found frame (2), and select four corners (,,,
) Is found, the position of the frame (2) on the paper is obtained based on the information of the four corners (,,,) of the above frame, and the pattern (3) of letters, numbers, symbols, etc. in the obtained frame (2) A frame cutting processing method in character recognition, which is characterized by cutting out. 2. The four corners of the found frame (,,,)
The slope of the frame (2) is calculated from the information of
2. The frame cutting processing method for character recognition according to claim 1, wherein the pattern (3) of characters, numbers, symbols and the like is cut out from the frame (2) which has been subjected to the rotation correction and which is subjected to the rotation correction.