JP2926842B2 - Character extraction circuit - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character recognition device, and more particularly to a character extraction circuit.

[Conventional technology]

Conventionally, this type of character extraction circuit uses a method of compressing the entire image in one direction (generally in a direction perpendicular to the image scanning direction) by inputting the image in binary, and extracting the image using a projected image of a character block. And a compression pattern register, a character block information creating / updating circuit, and a cut-out condition detecting circuit.

[Problems to be solved by the invention]

In the above-described conventional character extraction circuit, since a character block is detected by a compression pattern, a diagonal character (for example, FIG. 7)
Since such images cannot be detected separately, they often hinder the improvement of reading performance.

FIG. 8 is a diagram showing a process of processing the image of FIG. 7 by the conventional method. FIG. 8 (1) shows a processing state when the middle of the left character in the input image is scanned. Compression pattern created by the previous scan (A-1)
And the input image obtained by scanning the point A, the compression pattern at the point A is obtained. FIG. 8 (2) is a diagram showing a processing state when the rightmost end of the left character is scanned.
Originally, the left character is cut out when point B is scanned. However, since the character is described diagonally, the left end pattern of the right character is inserted. As a result, a compression pattern B is obtained, and the two characters are Because they are combined, two characters cannot be separated and cut out.

[Means for solving the problem]

The character cutout circuit according to the present invention includes an image memory for storing an image such as a form, an address control unit for scanning the image memory space, a feature buffer, and 3 × 3 mesh grid pixels. The connection of the black lattice of the center pixel and the peripheral pixels is coded by eight vectors toward the pixel, and the two coordinates of the lower left and the upper right of the lattice pixels are combined into a feature buffer as a characteristic of the 3 × 3 mesh lattice pixels. A pixel feature extraction unit to store, and every time the scanning of the input pixel moves by one pixel, the lower left coordinate of the connected grid pixel stored in the feature buffer is the most lower left coordinate of the connected grid pixel. Update to a smaller one, and change the upper right coordinates of the connected grid pixels to the largest of the upper right coordinates of the connected grid pixels. It has a diagonal character cutout unit that outputs the coordinates stored in the feature buffer as the final cutout information when the black pixels are updated and the continuous black pixels in the vector direction as the determination target for detecting the character chunk are no longer seen.

[Action]

According to the present invention, a correlation between pixels is extracted as a pixel feature without detecting a character by a compression pattern, and a character block is detected and extracted based on a connection between the features to obtain cutout information. Therefore, in the conventional method, the two patterns can be integrated by the compression pattern and the two patterns can be completely separated and cut out.

〔Example〕

Next, embodiments of the present invention will be described with reference to the drawings.

FIG. 1 is a block diagram of a character extracting circuit according to an embodiment of the present invention. FIG. 2 is a diagram showing a correlation between a central pixel and eight peripheral pixels surrounding the central pixel in a 3 × 3 mesh lattice pixel. FIG. 3 is a diagram showing an example of an input pixel and its features and the coordinates of a center pixel. FIG. 4 is a diagram showing an example of an input pixel and its features and coordinates of a center pixel. FIG. 6 is a diagram showing an example of input pixels, and FIG. 6 is a diagram showing lattice pixels at positions of X = A and X = B in the input pixels of FIG.

The pixel feature extraction unit 2 scans the image to be processed from bottom to top, and in a 3 × 3 mesh lattice pixel moving from left to right, the correlation between the center pixel and eight surrounding pixels surrounding it, that is, The degree of connection of the black lattice is coded, and is expressed as a pixel feature together with the coordinates of the center pixel. For simplicity, the correlation is represented by eight vectors af
(FIG. 2 (a)) represented by the coordinates of the _{X S, Y S, Y L} , Y L ( 3
(B) of FIG. These coordinates are sequentially updated by the connection of the lattice pixels, and are later used as cutout information. Now, when a pixel as shown in FIG. 2A is input, the center pixel and the black lattice are connected by the vectors e,
In the pixel of f, the coordinates of the center pixel are (X _S , Y _S , X _L , Y _L ) = (p
−1, q−1, p + 1, q−1). The same applies to the case of FIG. 3 (b). In the 3 × 3 mesh grid pixels, the oblique character extraction unit 4 includes the features of the center pixel obtained by the pixel feature extraction unit 2 and the features of the peripheral pixels previously stored in the feature buffer 3 by scanning up to the previous row. Then, the connection is confirmed, cut-out information is created / updated, and stored as a pixel feature in the feature buffer 3 again. Also, depending on the conditions, the created cutout information is output to the outside as final character cutout information.

Assuming that the input pixels are shown in FIG. 4, the characteristics of each of the lattice pixels A, B, and C are obtained as shown in FIG. 4B.
Now, when the pixel feature extraction process is performed by focusing on the grid pixel B, the connection with the surrounding pixels is viewed. The connection vector f indicated by the grid pixel A detected during the front row scanning and the grid pixel currently being processed are displayed. The connected vectors indicated by B face each other, indicating that the grid pixels A and B are continuous. The coordinate information of the features of the lattice pixel A and the coordinate information of the lattice pixel B stored in the feature buffer 3 in advance are integrated, and the cut-out information is updated. That is, by integrating the coordinate information, the coordinates XS,
For YS, the smaller of the coordinates of the grid pixels A and B, XS, YS, is obtained and updated to that value. For XL, YL, the larger of the coordinates of the grid pixels A, B, XL , YL is determined and updated to that value.

In the lattice pixel B, the black pixels connected in the vector d and e directions are not integrated at this time because the lattice pixels B and C are related to each other, and the lattice pixel B and the lattice pixel detected in the next column scan. Therefore, when the processing for the lattice pixel B is completed, the vector indicating the integrated pixel feature and the coordinate extraction information are as shown in FIG. 4 (c). Next, when processing is performed by focusing on the lattice pixel C, the continuity with the lattice pixels A and B is detected in the same manner.
All of the grid pixels A, B, and C are integrated. Therefore, the characteristics of the grid pixels A, B, and C after integration are as follows: grid pixel A: (e, f) (p-1, q-1, p + 2, q + 2) grid pixel B: (b, d, e) (P−1, q−1, p + 2, q + 2) Grid pixel C: (a, e, f, h) (p−1, q−1, p + 2, q + 2).

As described above, when continuity is detected in the processing on the 3 × 3 mesh grid pixels that scan the processed image, the characteristics of these pixels are continuously updated. This process
The process is continued until a cutout condition for separating and cutting out a character block is detected. Here, the case where no continuous black pixel is seen in the directions of the vectors d, e, f, and g is set as the extraction condition.

As shown in FIG. 5, the 3 × 3 mesh reaches the rightmost end of the character block, in this case “1”, and when the extraction processing reaches the final stage, the oblique character extraction unit 4 detects the final extraction condition, The coordinates obtained by the characteristics of the lattice pixel Y are output as final cutout information.

The above pixel feature extraction unit 2 and oblique character extraction unit 4
In the configuration shown in the figure, the operation is performed on the input image of FIG. 7 which cannot be cut out by the conventional method. FIG. 6 and Table 1 show the processing results when X = A, Y = B of the input image shown in FIG. 7 is scanned. Here, (m, n) indicates the lower left coordinate of the left pattern, and (p, q) indicates the lower left coordinate of the right pattern.
Also, when X = B, the left pattern sets the cutout condition.

From the feature table of Table 1, it can be seen that the two patterns integrated by the compression pattern in the conventional method are completely separated into two and cut out in the present embodiment.

〔The invention's effect〕

As described above, the present invention extracts the correlation of each pixel as a pixel feature, detects and extracts a character block based on the connection of the features, and obtains cutout information. There is an effect that two broken patterns can be completely separated and cut out.

[Brief description of the drawings]

FIG. 1 is a block diagram of a character extracting circuit according to an embodiment of the present invention. FIG. 2 is a diagram showing a correlation between a central pixel and eight peripheral pixels surrounding the central pixel in a 3 × 3 mesh lattice pixel. FIG. 3 is a diagram showing an example of an input pixel and its features and the coordinates of a center pixel. FIG. 4 is a diagram showing an example of an input pixel and its features and coordinates of a center pixel. FIG. 6 is a diagram showing an example of an input pixel, FIG. 6 is a diagram showing a grid pixel at the position of X = A, X = B in the input image of FIG. 7, FIG. 7 is a diagram showing an example of an input image of oblique characters, 8th
The figure shows the result of processing the input image of FIG. 7 by the conventional method. 1 ... Image memory, 2 ... Pixel feature extraction unit, 3 ... Feature buffer, 4 ... Diagonal character extraction unit, 5 ... Address control unit.

Claims (1)

(57) [Claims] 1. An image memory for storing an image of a form or the like, an address control unit for scanning an image memory space, a feature buffer, and 3 × 3 mesh grid pixels from the center coordinates to eight peripheral pixels. The connection of the center pixel and the black lattice of the peripheral pixel is coded by eight vectors, and the 3 × 3 coordinates are added to the two coordinates of the lower left and upper right of the lattice pixel.
A pixel feature extraction unit that stores the feature of the grid pixel of the mesh in the feature buffer; and each time the scanning of the input pixel moves by one pixel, the lower left coordinate of the connected grid pixel stored in the feature buffer is connected to the connection. Update the lower left coordinate of the connected grid pixel to the smallest one, update the upper right coordinate of the connected grid pixel to the largest of the upper right coordinates of the connected grid pixel, and detect the character block A character cutout circuit having a diagonal character cutout unit for outputting the coordinates stored in the feature buffer as final cutout information when black pixels continuous in the vector direction to be determined are no longer seen.