JPH0677270B2 - Character recognition device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION Technical Field of the Invention According to the present invention, an unknown character is optically read and is read in black and white.
The present invention relates to a character recognition device that digitizes a character pattern, extracts characteristics of an unknown character based on the character pattern, collates the character with a standard pattern, and recognizes the unknown character.

<Background of the Invention> As shown in FIG. 1, a conventional character recognition device has a reading head 3 and a CCD (Charged-Coupled Decoder) that constitute a character reading unit.
vice) 4 and A / D converter 5, the unknown character 2 on the form 1 is read optically by the head 3, and this is read by the CCD4.
Is converted into an electric signal and further converted into a digital signal by the A / D converter 5. This digital signal is given to the pre-processing circuit 6, and pre-processing necessary for character recognition, such as data binary noise removal, smoothing, and rounding, is performed. The pre-processing circuit 6 measures the line width of the character pattern and the binarization circuit 8 for binarizing the output of the A / D converter 5 and storing the character pattern in the image memory 7, as shown in FIG. A line width measuring circuit 9 for outputting a line width judging signal i, a threshold value deciding circuit 10 for variably setting the threshold value of the binarization processing based on the contents of the line width judging signal i, and a character pattern. Data of a plurality of constituent meshes is converted into mesh conversion data of one mesh under a predetermined conversion condition to generate an inspection pattern,
A rounding circuit 12 for assembling this on the image memory 11 is included. Here, the mesh means a pixel (hereinafter the same). The pattern to be inspected is input to the feature extraction circuit 14 of the character recognition unit 13, the features of the character, for example, the number of intersections, the number of end points,
The number of branch points, the number of loops, etc. are coded and extracted, and this characteristic code is collated by the dictionary collating circuit 15 with the characteristic code of the standard pattern stored in the dictionary 16 in advance. As a result, if the two match, the ID code of the standard pattern having the characteristic code is output, and if they do not match, the reprocessing command j is sent to the preprocessing circuit 6. When the reprocessing command j is issued, the threshold value determination circuit 10 of the preprocessing circuit 6 changes the threshold value of the binarization processing based on the content of the line width determination signal i to change the line width of the character pattern. After the correction, the same character recognition process from the feature extraction is performed again based on the corrected pattern. However, in the case of this type of conventional method, when the density difference at the boundary portion between the white background and the black background in the character pattern exceeds a certain level, the correction effect of the character pattern may not appear remarkably even if the threshold value is changed. . For example, depending on the type of writing instrument used to write characters on a form, the density difference becomes extremely large,
In such a case, re-execution of character recognition processing is meaningless,
It only reduces the recognition processing efficiency.

<Object of the Invention> The present invention improves the correction effect of a character pattern by adopting a method of performing line width correction of a character pattern in a stage after the binarization process when re-executing the character recognition process. An object of the present invention is to provide a character recognition device for the above.

<Structure and Effect of the Invention> The character recognition device of the present invention is a binarization unit that reads an unknown character and converts it into an electric signal, and binarizes the output of the character reading unit into black and white to generate a character pattern. A processing unit, a line width measuring unit that measures the line width of the character pattern generated by the binarization processing unit and outputs a line width determination signal indicating the size of the line width, and the character pattern Data having a mask scanning section for dividing the mask area into pattern areas and extracting pattern data for each divided area by mask scanning, and a memory for converting each pattern data extracted by the mask scanning section into data forming a pattern to be inspected The conversion unit and the characteristic of the unknown character are extracted from the pattern to be inspected composed of the output data of the data conversion unit, and the unknown character is recognized by comparing with the standard pattern and the unknown character is recognized. When a character cannot be recognized, a character recognition unit that outputs a reprocessing command is provided. Then, the data conversion unit determines the combination of the presence or absence of a reprocessing instruction from the character recognition unit, the line width determination signal output from the line width measurement unit, and the pattern data extracted by the mask scanning unit. The data is input as a memory address, and the data forming the pattern to be inspected is output from the corresponding memory address.

According to the present invention, the line width correction effect of the character pattern is remarkably exhibited even when the density difference at the boundary portion between the white background and the black background in the character pattern is large. Therefore, the correction effect does not depend on the type of writing instrument used, the re-execution of the character recognition process becomes effective, and the character recognition accuracy is improved. .

<Description of Embodiments> FIG. 3 shows a partial structure of a character recognition device embodying the present invention.

In the figure, an A / D converter 5 converts the CCD output into digital signals of a plurality of levels and sends it to the preprocessing circuit 6. The pre-processing circuit 6 outputs the output of the A / D converter 5 at a constant level threshold value of 2
A binarization circuit 8 for binarizing and storing the character pattern in the image memory 7 (for example, a vertical and horizontal 64 mesh); a threshold value determining circuit 10 for setting the threshold value and giving it to the binarization circuit 8;
A line width measuring circuit 9 that measures the line width of a character pattern and outputs a line width determination signal i indicating the size of the line width, and a mesh of data of a plurality of meshes forming the character pattern according to a predetermined conversion condition. And a rounding processing circuit 20 for assembling the pattern to be inspected into the image memory 11 (for example, 32 meshes in the vertical and horizontal directions).

The rounding processing circuit 20 in this embodiment, as shown in FIG. 5, has a pattern data group a ₁ , a ₂ , a ₃ , ... For 4 meshes (2 meshes in the vertical direction × 2 meshes in the horizontal direction) which form a character pattern.
..., an, ... is the mesh conversion data of 1 mesh b ₁ , b ₂ ,
b _3, ..., bn, a circuit for converting data into ....,
As shown in FIG. 4, a mask scanning circuit 21 for sequentially scanning a pattern data group for four meshes by scanning the image memory 7 and an output of the mask scanning circuit 21 as address data, and storing the mesh conversion data at a corresponding address. It is composed of a read only memory (ROM) 22. This ROM
The address 22 is also defined by the line width determination signal i and the reprocessing command j from the character recognition unit. The presence or absence of the reprocessing command j, the line width determination signal i, and the mask scanning circuit 21 output the address. Based on the address data input consisting of the combination with the 4 mesh pattern data group, the mesh conversion data of any corresponding address is selectively output from the ROM 22, and the pattern to be inspected is assembled in the image memory 11.

For example, during normal processing, that is, when the reprocessing command j is not issued, as shown in FIG. 6 (1), the 4-mesh pattern data group includes black background data of 2 meshes or more (indicated by diagonal lines in the drawing). When, 1 as mesh conversion data
The black background data of the mesh (other than that, the white background data of 1 mesh) is set. That is, when the reprocessing command j is off and 2 or more meshes out of 4 meshes are black background data, the output data from the ROM 22 becomes black background data regardless of the content of the line width determination signal i. The mesh conversion data is set. Further, after the reprocessing, that is, when the reprocessing command j is issued, the conversion condition is changed according to the content of the line width determination signal i, and the content indicating that the line width determination signal i is "thick" 6 (2), when the 4-mesh pattern data group includes 3 or more black background data, 1 mesh black background data (other than that, 1 mesh white background data) as mesh conversion data. ) Is set, and the character pattern is thinned by this. That is, when the reprocessing command j is ON, and 3 or more meshes out of 4 meshes are black background data and the line width determination signal i has a content that means "thick", the output data from the ROM 22 is black background data. The mesh conversion data of the ROM 22 is set so that Further, at the time of reprocessing, that is, when the reprocessing command j is issued, if the line width determination signal i has a content that means "thinning", as shown in FIG. When the group includes one or more mesh black background data, one mesh black background data (other than that, one mesh white background data) is set as the mesh conversion data, and the character pattern is thickened. That is, when the reprocessing command j is on, 1 or more of the 4 meshes are black background data, and the line width determination signal i has a content that means "narrow", the output data from the ROM 22 is black background data. The mesh conversion data of the ROM 22 is set so that

Thus, when the unknown character cannot be recognized, the character recognition unit sends the reprocessing command j to the preprocessing circuit 6, and the rounding processing circuit 20 performs the thickening processing when the line width of the character pattern is thin. When it is thick, the thinning process is executed to reassemble the pattern to be inspected, the feature of the character is re-extracted from the pattern to be inspected, and the character recognition process is executed.

[Brief description of drawings]

FIG. 1 is a block diagram showing the overall configuration of the character recognition device, FIG. 2 is a block diagram showing the configuration of a conventional preprocessing circuit, and FIG.
FIG. 4 is a block diagram showing a configuration of a preprocessing circuit according to the present invention, FIG. 4 is a block diagram showing a configuration of a rounding processing circuit, FIG. 5 is an explanatory diagram showing a mesh conversion process, and FIG. 6 is a mesh conversion pattern. It is the explanatory view illustrated. 6 ... Pre-processing circuit, 9 ... Line width measuring circuit 13 ... Character recognition unit, 20 ... Rounding processing circuit 22 ... ROM

Claims (1)

[Claims] 1. A character reading unit for reading an unknown character and converting it into an electric signal, a binarization processing unit for binarizing an output of the character reading unit in black and white to generate a character pattern, and the binarization processing unit. A line width measurement unit that measures the line width of the character pattern generated in step S1 and outputs a line width determination signal that indicates the size of the line width, and divides the character pattern into predetermined mask areas and A mask scanning section for taking out pattern data by mask scanning, a data converting section having a memory for converting each pattern data taken out by the mask scanning section into data forming a pattern to be inspected, and an output of the data converting section The characteristics of unknown characters are extracted from the pattern to be inspected that is composed of data, and the unknown characters are recognized by matching them with the standard pattern. And a character recognition unit for outputting a command, the data conversion unit, the presence or absence of a reprocessing command from the character recognition unit, a line width determination signal output from the line width measurement unit, the mask scanning A character recognition device configured to input a combination with pattern data extracted by a unit as an address of the memory, and to output data forming a pattern to be inspected from a corresponding memory address.