JPH07192097A - Character recognition device and method therefor - Google Patents

Abstract

PURPOSE:To recognize characters without being influenced by the deviation of the characters. CONSTITUTION:This character recognition device is provided with an A/D conversion circuit 2 for digital-converting the output video signals (a) of a photoelectric conversion scanner 1 for picking up the images of recognition object character pictures and outputting digital pictures (b), a character data grid division circuit 5 for outputting the output character data (d) of a character peripheral area segmenting circuit 4 for segmenting a character area and a character peripheral area from the output picture data (c) of a picture memory 3 for tentatively storing the digital pictures (b) as grid division data (e) which are neural net recognition input patterns, a neural net character recognition circuit 7 for character-recognizing the output shift data (f) of a character data shifting circuit 6 prepared by gathering the data for which the grid division data (e) are shifted in eight upward and dounward, leftward and rightword and oblique directions and the data for which the grid division data (e) are not shifted by using a neural net and outputting a judgement output value (g) and a recognized result judgement circuit for defining the character whose neural net judgement output value (g) becomes maximum as a recognized character (h).

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character recognition device and a method thereof, and more particularly to a character recognition device and a method thereof which can obtain a high recognition rate without being influenced by noise even when a character is cut out by shifting the character.

[0002]

2. Description of the Related Art Conventionally, a character recognition device of this type has learned character data that has been shifted in advance when learning a character, and recognizes character data that has been misaligned during recognition. Even if the recognition target character is cut out due to noises such as the characters in FIG. 5, the recognition is still performed.

[0003]

Since this conventional character recognition device recognizes the character data cut out as it is, the character "E" is shifted downward as shown in FIG. It is not possible to identify whether "" has shifted downwards or "F" has shifted downwards, and there is a possibility that it may be erroneously recognized as "F", which causes a reduction in the recognition rate.

[0004]

A character recognition apparatus of the present invention is an image pickup means for picking up an image of a character to be recognized and outputting a video signal, and an A / A for digitally converting the video signal and outputting a digital image. D conversion means, image storage means for temporarily storing the digital image and outputting it as image data, and character peripheral area cutout means for cutting out a character area and a character peripheral area from the image data and outputting it as cutout character data, Character data grid division means for outputting the cut-out character data as grid division data which is a neural network recognition input pattern, and 9 data which is a combination of data shifted from the grid division data in eight directions, and data not shifted. The character data shifting means to be created and the 9 data are respectively recognized by using a neural network. It comprises a neural network character recognition means for outputting a neural net decision output value, and a recognition result determining unit that the neural net decision output value is the recognized character text of maximum.

[0005]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Next, the present invention will be described with reference to the drawings.

Referring to FIG. 1 which is a block diagram showing an embodiment of the present invention, a character recognition apparatus according to this embodiment includes a photoelectric conversion scanner 1 for picking up an image of a character to be recognized and outputting a video signal a; An A / D conversion circuit 2 that digitally converts the signal a and outputs a digital image b, and an image memory 3 that temporarily stores the digital image b and outputs it as image data c.
And a character peripheral area cutout circuit 4 which cuts out a character area and a character peripheral area from the image data c and outputs it as cutout character data d, and outputs the cutout character data d as grid division data e which is a neural network recognition input pattern. The character data grid dividing circuit 5, the character data shifting circuit 6 that creates 9 data, which is data obtained by licking the grid divided data e in the eight directions and the data that is not shifted, as shift data f, and the shift data f
A neural network character recognition circuit 7 for recognizing characters using a neural network and outputting a neural network judgment output value g, and a recognition result judgment circuit for recognizing a character having the maximum neural net judgment output value g as a recognition character h. Composed of.

FIG. 2 is a diagram showing an example of cut-out character data d obtained by cutting out the image data c by the character peripheral area cut-out circuit 4. The cut-out character data d in FIG. 2 is
The character peripheral area 21 is expanded by vertically and horizontally by one grid when dividing the noisy character area 22 of the character "E" into a grid which is a neural network recognition input pattern in consideration of later shifting. It is the cut-out image data.

FIG. 3 shows an example of grid division data e obtained by dividing the cut-out character data d of FIG. 2 by the character data grid division circuit 5. The grid division data e in FIG. 3 is divided into 9 × 11 grids as a neural network recognition input pattern.
X13. In the grid division data e, the size of the character portion (black portion) in one grid is represented by the size of the black square in the white cell.

FIG. 4 shows an example of shift data f created by the character shift circuit 6. The character data shift circuit 6
9 pieces of data 41 to 48 obtained by shifting the grid division data e by one grid in the vertical and horizontal licking directions and the non-shifting data 40 are created and output as shift data f.

The neural network character recognition circuit 7 character-recognizes the shift data f of the 9 data 40 to 48 by using a neural network, respectively, and judges with the value of 0 to 1.0 indicating the certainty of the recognized character. Output the output value g. Judgment output value g in the case of the unshifted data 40 in FIG.
, The first candidate is the letter'F ', 0.312, and the second candidate is the letter'E', 0.207. In the case of the upper shift data 41, the determination output value g is 0.987 for the character “E” for the first candidate and 0.010 for the character “3” for the second candidate. Among the shift data f of these 9 data 40 to 48, the upper shift data 41 is the closest to the data to be originally cut out, so the judgment output value g of'E 'becomes high, and the other 8 data is the output of the first complement. The value does not become that high either.

The recognition result judging circuit 8 is composed of 9 data 40-4.
The character having the maximum output value is determined as the recognition character h from the determination output value g of the shift data f of 8, and “E” is output as the recognition character h.

[0012]

As described above, according to the present invention,
A character area and a peripheral area of the character are cut out from the image data and output as grid division data which is a neural net recognition input pattern, and the grid division data is combined with data that is shifted in eight directions and is not shifted. It is possible to perform character recognition that is not affected by character misalignment by creating 9 data and character-recognizing each of these 9 data using a neural network and setting the character with the maximum neural network determination output value as the recognized character. it can.

[Brief description of drawings]

FIG. 1 is a block diagram showing the configuration of an embodiment of the present invention.

FIG. 2 is a diagram showing an example of cut-out character data of this embodiment.

FIG. 3 is a diagram showing an example of grid division data of this embodiment.

FIG. 4 is a diagram showing an example of shift data of this embodiment.

FIG. 5 is a diagram showing a case where noise is misaligned and cut out in a character to be recognized.

[Explanation of symbols]

 1 Photoelectric conversion scanner 2 A / D conversion circuit 3 Image memory 4 Character peripheral area cutout circuit 5 Character data grid division circuit 6 Character data shift circuit 7 Neural net character recognition circuit 8 Recognition result judgment circuit 21 Character peripheral area 22 Character area a Video Signal b Digital image c Image data d Cutout character data e Lattice division data f Shift data g Judgment output value h Recognition character

Claims (2)

[Claims] 1. An image pickup means for picking up an image of a character to be recognized and outputting a video signal, an A / D conversion means for digitally converting the video signal to output a digital image, and temporarily storing the digital image. Image storage means for outputting the image data as image data, a character peripheral area cutting-out means for cutting out a character area and a character peripheral area from the image data and outputting it as cut-out character data, and the cut-out character data as a neural network recognition input pattern. The character data grid dividing means for outputting as grid division data, the character data shifting means for creating 9 data including the data shifted from the grid division data in the eight directions of up, down, left and right, and the data not shifted, and the 9 data Characters are recognized using a neural network and the neural network judgment output value is output. A character recognition device comprising: a Uralnet character recognition unit; and a recognition result judgment unit that recognizes a character having the maximum neural net judgment output value as a recognition character. 2. An image of a character to be recognized is captured, a video signal is output, the video signal is digitally converted into a digital image, the digital image is temporarily stored and output as image data, and the character is converted from the image data. Data obtained by cutting out a region and a peripheral region of a character and outputting it as cut-out character data, outputting the cut-out character data as grid division data that is a neural network recognition input pattern, and shifting the grid division data in eight directions in vertical and horizontal directions. Nine data are created by combining the non-shifted data, the nine data are respectively subjected to character recognition using a neural network, and a neural net determination output value is output, and a character having the maximum neural network determination output value is recognized. A character recognition method characterized by using characters.