JPS603679B2 - Character recognition method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a character recognition method that extracts outline information of characters and performs character recognition based on the extracted outline information.

In conventional character recognition methods, characters are subjected to line thinning processing, and character features are extracted from the thinned character information.

The character recognition method that involves line thinning processing as described above has the drawback that the line thinning process requires a very large number of processing steps. The present invention aims to eliminate the above-mentioned drawbacks, and in order to reduce the number of processing steps required for character feature extraction, the present invention provides a character recognition method that performs character recognition by extracting character contour information. is intended to provide.

Therefore, the character recognition method of the present invention includes an image memory in which quantized image information is written and a scanning means for scanning the image memory.
Every time one line of the image memory is scanned, the first appearing change point from white to black or the last appearing change point from black to white is detected, and the distance between the changing points in the adjacent scanning line exceeds a predetermined value. Based on the detected discontinuity, one or more points defined by the start point and end point are detected. an outer contour line information extraction means for extracting information regarding an outer contour line consisting of segments; and a point of change from black to white and a change from white to black following the point of change each time one line of the image memory is scanned. If there is a white area sandwiched between black areas in the current scanning line and there is no white area of this kind that is in contact with the white area in the previous scanning line, the black area in the current scanning line is detected. The change points at each end of the white area sandwiched between are defined as feature points, and if there was a white area sandwiched between black areas in the previous scan line, and this type of white area that is in contact with the white area is currently being scanned. If the white area does not exist in a row, the points of change in the white area in the previous scanning line are extracted as feature points, and based on the obtained feature points, information about the outline of the white area surrounded by black is extracted. The present invention is characterized in that a contour information extraction means is provided, and character recognition is performed based on the above-mentioned outer contour information or white portion contour information. Hereinafter, the present invention will be explained with reference to the drawings. 1 to 3 are diagrams for explaining the principle of contour information extraction according to the present invention, and Figure 4 is a diagram for explaining the principle of contour information extraction according to the present invention.
7 to 7 are diagrams for explaining examples of character recognition based on outline information, and FIG. 8 is a block diagram of one embodiment of the present invention.

FIG. 1 explains the first principle of contour information extraction employed in the present invention.

In FIG. 1, S represents a start point, E represents an end point, and 0 represents a flag. Now, the entire area of the figure in FIG. 1 is scanned with a predetermined scanning line, and the first appearing point of change from white to black is plotted. In the change point series obtained in this way, when the distance in the scanning direction between adjacent change points is a certain threshold value, for example, 3 mesh points or more, it is defined as discontinuous, and several segmented changes are thereby defined. Obtain a point series. The left end of each segment is defined as a start point S, and the right end is defined as an end point E.
Then, if the direction from end point E to start point S is equal to the scanning line direction, a flag is attached to start point S, and if the direction from end point E to start point S is opposite to the scanning line direction, a flag is attached to start point S. Add a flag to endpoint E. Figure 1 shows the start point S, end point E, and flag when a figure is scanned from left to right with a scanning line going from top to bottom, and the figure is scanned from left to right with a scanning line going from bottom to top. It shows the start point S, end point E, and flag in the case. Lower contour information can also be obtained from a series of black-to-white transition points that appear last when scanning with a scanning line going from top to bottom. In this case, when the direction from the end point to the start point is equal to the scanning direction, a flag is attached to the end point, and when the direction from the end point to the start point is opposite to the scanning direction, a flag is attached to the start point. It goes without saying that the start point S, end point E, and flag are formed in the same way even when a figure is scanned with a scanning line going from left to right or from right to left. The start point S, end point E, and flag constitute outer contour information. FIG. 2 shows the second principle of contour information extraction employed in the present invention. In FIG. 2, RU indicates the upper right point, RL the lower right point, LU the upper left point, LL the lower left point, and 0 a flag. Figure 2 shows the upper left point LU, lower left point LL, and upper right point R generated when a figure is scanned from left to right with a scanning line going from top to bottom.
U, the lower right point RL and a flag are shown. In a white area surrounded by black on the top and bottom, when scanning from top to bottom, a point of change from black to white appears first, and then a point of change from white to black appears. The point of change from black to white at the left end of the white portion is called the upper left point LU, and the point of change from white to black is called the lower left point LL. Also, the point where the right end of the white part changes from black to white is the upper right point RU
The point of change from white to black is called the lower right point RL. If black exists to the left of the left end upper point LU or the left end lower point LL, a flag is attached to the left end upper point LU or the left end lower point LL'. In addition, if black exists on the right side of the upper right point RU or the lower right point RL, the upper right point RU or the lower right point R
A flag is attached to L. These left end upper point LU, left lower point LL, right upper point RU, right lower point RL, and flag constitute outline information of a white portion surrounded by black. When the entire figure is scanned with a scanning line going from left to right, outline information of a white part surrounded by black can be obtained in the same way. Figure 3 shows the upper left point LU, the lower left point LL, the upper right point RU, and the lower right point RL when the white part surrounded by black has branches.
and flags.

The fourth example shows an example of identifying the character "o" and the character "sa" based on outline information.

For the letter "o", RL-LL RU-LU is much larger than 1, and for the letter "sa", RL-
LLRU-LU is approximately 1.

The character "o" and the character "sa" can be distinguished by using these differences in properties. FIG. 5 shows an example of identification of the number "2" and the number "3" based on contour information.

The number "2" has one upper right point RU, but the number "3" has two upper right points RU. The number "2" and the number "3" can be distinguished by utilizing such a difference in properties. Figure 6 shows the character ``power'' and the character ``power''.
This example shows an example of how to identify a camera.

In the character "chi", the x-direction line segment of the difference between the end point ■ and the start point S is negative, and in the character "cha", the x-direction component of the difference is positive. Utilizing these differences in properties, it is possible to distinguish between the character ``chi'' and the character ``ya''. FIG. 7 shows an example of identifying the letter "S" and the number "5".

For the letter "S", the difference between the end point E obtained when scanning with a horizontal scanning line going from right to left and the starting point S obtained when scanning with a vertical scanning line going from top to bottom. The X direction component is greater than or equal to the predetermined value n. On the other hand, for the number "5", the end point E obtained when scanning with a horizontal scanning line going from left to right and the starting point S obtained when scanning with a vertical scanning line going from top to bottom. The difference in the x-direction line segment is less than or equal to the predetermined value n. Taking advantage of these differences in properties,
Can distinguish between the letter "S" and the number "5". 8th
The figure is a block diagram of one embodiment of the present invention, in which 1 is a form, 2 is a sensor, 3 is a quantization circuit, 4 is an image memory, 5 is a contour address extraction circuit, 6 is a contour address memory, 7
-1 to 7-4 are start point/end point extraction circuits, 8-1 and 8-2 are white part continuous feature extraction circuits surrounded by black, 9-1 to 9-4 are start/end feature pair memories, 10-1 and 10-2 are white portions surrounded by black feature pair memories, and 11 is a judgment calculation/logic discrimination circuit, respectively.

The light from the form 1 is converted into an electrical signal by the sensor 2, this electrical signal is input to the quantization circuit 3 and converted to binary image information, and this binary image information is stored in the image memory 4. .

The entire area of the image memory 4 is scanned with vertical scanning lines going from top to bottom, and then scanned from top to bottom with horizontal scanning lines going from left to right. The position of the change point from white to black, the position of the change point from white to black, and the current scan line position in the current scan line are extracted by the contour address extraction circuit 5, and the position of these change points and the current scan line position are used as the contour address. -Written to memory 6. When scanning the image memory 4 with a vertical scanning line going from top to bottom, the position of the change from white to black that first appears in the current scanning line and the current scanning line position are sent to the start point/end point extraction circuit 7-1. , the position of the change point from black to white that appears last, and the current scanning line position are sent to the start point/end point extraction circuit 7-2. Further, the position of the change point from black to white in the current scan line, the position of the change point from white to black that appears next, and the continuous feature extraction circuit 8-
Sent to 1. The start point/end point extraction circuit 7-1 compares the change point position in the scan line one scan before with the change point position in the current scan line, and if the two are discontinuous, the change point position in the scan line one scan before is compared. The position of the change point is taken as an end point, the position of the change point in the current scanning line is taken as a start point, and a flag is attached according to a predetermined rule, the corresponding start point and end point are paired, and the start-end feature pair memory 9-1 write to.

The start point/end point extraction circuit 7-2 also performs a similar operation. If there is a white area sandwiched between black areas in the current scanning line and there is no white area of this kind that is in contact with the white area in the previous scanning line, the white part continuous feature extraction circuit 8-1 extracts the white area from the current scanning line. The ends of the white area sandwiched between the black areas in are defined as the upper left point LU and the lower left point LL, respectively, and these are written as a pair in the white portion feature pair memory 10-1.

In addition, the white part continuous feature extraction circuit 8-1 detects when there is a white area sandwiched between black areas in the previous scanning line and there is no white area of this type that is in contact with the white area in the current scanning line. , each end of the white area in the scanning line one scan before is set as the upper right point RU and the lower right edge RL, and these are written as a pair in the white portion feature pair memory 10-1. At this time, flags are attached according to the rules. The start point/end point extraction circuits 7-3, 7-4 and the white portion continuous feature extraction circuit 8-2 are used when the image memory 4 is scanned by horizontal scanning lines.

Since these operations are likely to be understood from the previous explanation, further explanation will be omitted. The determination calculation/logic identification circuit performs character recognition from the outline information of the start/end feature pair memories 9-1 to 9-4 and the white portion feature pair memories 10-1 and 10-2.

Several examples of such recognition logic are described in FIGS. 4-7.
As is clear from the above description, according to the present invention, it is possible to obtain contour line information that accurately represents a character by scanning the entire image memory once or twice. In comparison, the number of processing steps can be significantly reduced.

[Brief explanation of the drawing]

Figures 1 to 3 are diagrams for explaining the principle of contour line extraction according to the present invention, Figures 4 to 7 are diagrams for explaining an example of character recognition using contour line information, and Figure 8 is a diagram for explaining an example of character recognition using contour line information. 1 is a block diagram of one embodiment of the present invention. FIG. 1...Form, 2...Sensor, 3...Quantization circuit, 4...
... Image memory, 5 ... Contour address extraction circuit, 6 ... Contour address memory, 7-1 None,
7-4...start point/end point extraction circuit, 8-1 and 8-2...white part surrounded by black continuous feature extraction circuit, 9-1 to 9-4... ...Start-end feature vs. memory, 10-1 and 10-2...
White part surrounded by black feature pair memory, 11... Judgment calculation/logic identification circuit. Spear thigh 2 evil 3 female na 4 evil spear S figure na G Shichichiri evil 8 evil

Claims (1)

[Claims] 1. In a character recognition system that includes an image memory in which quantized image information is written and a scanning means that scans the image memory, the first white to black character that appears each time one line of the image memory is scanned is Discontinuity in a point series consisting of multiple change points is detected by detecting a change point or the last change point from black to white, and determining whether the distance between the change points in adjacent scanning lines exceeds a predetermined value. detect the part,
an outer contour information extraction means for extracting information regarding an outer contour line consisting of one or more segments defined by a start point and an end point, based on the detected discontinuous portion; and Every time one line of the memory is scanned, a point of change from black to white and a point of change from white to black that follows the change point are detected. If this type of white area adjacent to the area does not exist in the previous scan line,
The change points at each end of the white area sandwiched between the black areas in the current scanning line are defined as feature points, and the white area sandwiched between the black areas in the previous scanning line exists and this kind If the white area does not exist in the current scanning line, the points of change in the white area in the previous scanning line are extracted as feature points, and based on the obtained feature points, the outline of the white area surrounded by black is extracted. 1. A character recognition method, comprising: white part contour information extraction means for extracting information, and character recognition is performed using the outer contour information or white part contour information.