JPS59189481A - Character recognizing device - Google Patents

Abstract

PURPOSE:To improve efficiency of processing of character recognition and to make speed of character recognition constant by counting the number of profile black picture elements while judging whether each constituting picture element is black picture element or not and whether the picture element is profile black picture element or not in scanning a character picture. CONSTITUTION:An unknown character 2 indicated on a sheet of paper 1 is read by a pattern reading device 3 consisting of CCD etc. and its optical pattern is converted to a digital signal of time series by an A/D converter 4. A preprocessing circuit 5 performs removing of noise of the digital signal and binary coding by a specified threshold level, and stores a character picture in a picture memory 6. The picture memory 6 is made up of plural vertical and horizontal picture elements, and picture data of binary coded character picture are stored in each picture element. A dictionary collating circuit 8 collates characteristic of character picture obtained by a characteristic extracting circuit 7 with examples of characteristic of reference pattern stored beforehand in a dictionary 9, and outputs the result of collation.

Description

[Detailed Description of the Invention] <Technical Field of the Invention> The present invention involves reading an optical pattern of unknown characters, symbols, etc. (hereinafter referred to as "unknown characters"), converting the optical pattern into black and white to obtain a character image, and then The present invention relates to a character recognition device that recognizes unknown characters by comparing character images with standard patterns. The present invention relates to a character recognition device having a function of determining and adjusting the line width of a character by comparing the total number of black pixels with the total number of black pixels forming an outline (hereinafter referred to as "contour black pixels").

<Background of the Invention> In a conventional character recognition device, when the optical pattern of an unknown character is binarized, the line width of the character image changes depending on the size of the thread taco 4 hold level, and this causes the corresponding standard, N6 The results of the comparison with Kuhn may not match, and it may be determined that the object is unrecognizable. In this case, the threshold level is changed to adjust the line width of the characters, and then the pattern matching process is executed again. Generally, the method of determining the line width of a character is to find the total number of black pixels of the character image and the total number of outline black pixels, then calculate the ratio of the two, and determine the line width of the character image based on the size of the ratio. method has been adopted. In order to calculate the total number of outline black pixels, conventionally, the constituent pixels of the image memory in which the character image is stored are sequentially scanned, and the black pixels are detected and the black pixels forming the character outline are detected. In this case, the contour line is traced to count the number of contour black pixels, and the same tracing process is performed for other contour lines, and then the total number of contour black pixels is calculated for all contour lines. Therefore, with this type of method, it takes a significant amount of time to calculate the total number of black contour pixels, which greatly reduces the processing efficiency of character recognition.Also, since the number of contour lines differs for each character, the calculation of the total number of black contour pixels There were problems such as the time fluctuating and the character recognition processing rate being inconsistent.

<Object of the Invention> The present invention eliminates the method of tracing the outline of a character, and determines the number of outline black pixels by determining whether each constituent pixel is a black pixel in scanning a character image, and further determining whether or not it is an outline black pixel. It is an object of the present invention to improve the processing efficiency of character recognition by counting , and to provide a new character recognition device in which the character recognition speed does not fluctuate.

<Structure and Effects of the Invention> In order to achieve the above object, the present invention scans an image data storage means storing black and white binary image data with a mask having a fixed field of view, and performs black and white detection for the central pixel within the mask. The black pixel detection signal and the contour black pixel detection signal are obtained by performing discrimination and contour line discrimination, and these signals are counted to calculate the total number of black pixels and the total number of contour black pixels of the character image.

According to the present invention, the total number of black pixels and the total number of contour black pixels can be determined by sequentially scanning the image data storage means, so that the line of the character It is possible to significantly reduce the processing time for width determination and character recognition, and also to prevent fluctuations in the recognition processing time for each character, achieving the excellent effects of achieving the purpose of the invention.

<Explanation of Examples> Figure 1 shows the overall configuration of a character recognition device.

In the illustrated example, unknown character 2 displayed on page 1 is CC
It is read with a pattern reading device 3 consisting of D, etc.
The optical pattern is converted into a time-series digital signal by A/D converter No. 4. The preprocessing circuit 5 removes noise from the digital signal and further performs binarization processing using a predetermined threshold level, and writes the character image into the image memory 6. The image memory 6 is composed of a plurality of pixels in the vertical and horizontal directions, and image data of a binary character image is stored in each pixel. The feature extraction circuit 7 extracts several features (intersections, branch points, number of loops, stroke length, etc.) necessary for character recognition from the character images stored in the image memory 6.

The dictionary matching circuit 8 matches the features of the character image obtained by the feature extraction circuit 7 with feature examples of standard patterns stored in advance in the dictionary 9, and outputs the matching result. If the character cannot be recognized due to the line width of the character, the threshold level is changed and the same process and procedure as above is executed again. In the figure, CPU (CenLral Process
ssing Unit) l Q is RAM (R
andam Access Memory ) l l
It reads and writes various data necessary for character recognition processing, controls the operation of each of the circuits mentioned above, and executes various calculations and processes.

FIG. 2 shows a character line width determination circuit. 3 RAs in the diagram
M12A, 12B, and 12C are for capturing three lines of image data stored in the image memory 6, and the image data for three pixels in each of ΔM12A to 12G are stored in shift registers 13A, 13B, and 13G, respectively. is set to

Counters 14A, 14B, ILC are RAM12
'A to 12C are used for address setting, and image data is sequentially read out from each RAM 12A to 12C in accordance with the counting operation, and the shift registers 13A to 13C are simultaneously shifted to the right. Therefore, each shift register 3
Image data for 3 pixels each in 3 rows is set in A to 13C, and by the simultaneous shift operation of each shift register 3A to 1' 3C, a *mask W (No. 3) to obtain the same effect as scanning the constituent pixels of the pixel memory 6 in the row direction.

Each bit output of each shift register 3A to 13C is sent to a table ROM (Read
0nly Memory) 15. This table ROM 15 is used to determine whether or not the center pixel e of the mask (a) is a black pixel, and furthermore, whether or not it is a contour black pixel, and based on the pixel determination data, the black pixel detection signal i and the contour black pixel detection signal J Output.

FIG. 4 is a diagram explaining the principle of determining black pixels on the outline, and it is assumed that the central pixel e in the mask W is a black pixel, and 1 to 3 pixels among the four pixels s, d, [, h When the pixel is a black pixel, the center pixel C is assumed to be an outline black pixel. Therefore, in this embodiment, the table RO
The address of M15 (9 bits) is defined, and the address of the bit configuration corresponding to the black pixel array is data indicating that it is an outline black pixel, and the address of the bit array where the center pixel e is a black pixel is black. Data indicating that each pixel is a pixel is stored. For example, if we assume that the shaded area in Figure 4 is a black pixel, then the address "xQX 011 XIX J (where X
(can be either "1" or "0") stores data indicating that the pixel is a black pixel or a pixel with a black outline.

In this way, the table ROM 15 outputs either a black pixel detection signal i based on the black pixel data or a contour black pixel detection signal j based on the contour black pixel data, and each of these signals '+
The number of signals of J is counted by counters 16' and 17, respectively.

The outputs of each counter 16, 17 are connected to the CPUIQ data bus, and when scanning of the image memory 6 is completed, the
PUIQ I/. The final count data N1+NZ of each counter 16, 17 is sent to cpul via the data bus according to the instruction.
It is taken into o. One of the final count data N1 is the number of outline black pixels in the character image, and the other final count data N1
2 corresponds to the number of black pixels, and CPUIQ is the ratio of both 5-N
1/N2 is calculated, and based on the size of the ratio S, it is determined whether the line width of the character is thick or thin. For example, in the case of the character image for the English letter r FJ shown in FIG. 5, by scanning it with a mask W, it is possible to detect outline black pixels indicated by ○ marks in the figure and other black pixels indicated by X marks. Furthermore, the line width of the character is determined by determining the ratio between the number of pixels of the circle mark and the number of pixels of the circle mark and the X mark.

However, if the line width of the character is thick, the threshold level of the preprocessing circuit 5 is reset to a higher value, while if the line width of the character is thinner, the threshold level is reset to a lower value. The line width of the stored character image is adjusted and the same character matching process is executed again.

[Brief explanation of drawings]

Figure 1 is a block diagram showing the overall circuit configuration of the character recognition device, Figure 2 is a block diagram showing the character line width determination circuit, and Figure 3 is a block diagram showing the character line width determination circuit.
The figure is a diagram for explaining the mask set for the image memory, Figure 4 is a diagram showing the pixel range that falls within the field of view of the mask, and Figure 2 is a diagram showing outline black pixels and other black pixels in an example of a character image. It is a figure explaining arrangement of. 6... Image memory 10... CPU
13A, 13B, 13C...Shift register 15...Table ROM 16.17...Counter W...Mask patent applicant Tateishi Electric Co., Ltd. ,,ae Monument 2

Claims (1)

[Claims] In a character recognition device that has a function of determining the line width of a character by comparing the total number of black pixels of a character image and the total number of outline black pixels,
An image data storage means for storing black and white binarized image data in a plurality of pixels; scanning detection means for outputting a black pixel detection signal and a contour black pixel detection signal when executed; and a counting means for counting each signal output of the scanning detection means to calculate the total number of black pixels and the total number of contour black pixels of the character image. A character recognition device comprising: