JPH03253987A - Picture normalizing system - Google Patents

Abstract

PURPOSE:To shorten the processing time by determining a point on a normalized picture corresponding to each picture element on the normalized picture in accordance with a ratio of sizes of the picture before and after normalization and determining a black or white picture element correspondingly to the picture element of the point on the unnormalized picture corresponding to a noticed picture element on the normalized picture. CONSTITUTION:A corresponding point determining part 12 determines coordinates of a corresponding point in accordance with longitudinal and transverse sizes of the unnormalized picture stored in a picture memory 10 and preliminarily designated longitudinal and transverse sizes of the normalized picture and reports it to a converting part 14, and this part 14 reads out picture element information from the picture memory 10 with respect to the corresponding point of the normalized picture element in accordance with corresponding point coordinate information reported from the corresponding point determining part 12 and determines the color of the normalized picture element to write information '0' or '1' in a picture memory 16. Thus, high-speed normalization processing is possible.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to an image normalization method, and particularly to a normalization method suitable for normalizing a character image prior to feature extraction in a character recognition device.

[Conventional technology]

Generally, in a character recognition device, after normalizing a character image into an image having a certain specific height and width, feature quantities of the character image are extracted and matching processing with a dictionary is performed.

Conventionally, the following methods are known as such image normalization methods.

■ Normalization by OR (logical sum operation) For each image on the normalized image, find the pair J area on the pre-normalized image, and if there is even one black pixel in the corresponding area, it is the normalized pixel. is a black pixel.

■ Normalization by AND (logical product operation) Find the corresponding area on the pre-normalized image for each pixel on the normalized image, and if all pixels in the corresponding area are black pixels, consider the normalized pixel as a black pixel. do.

■ Normalization using a threshold This is a method described in Japanese Patent Application Laid-Open No. 52-43312, etc., in which a corresponding area on the pre-normalized image is determined for each pixel on the normalized image, and black pixels in the corresponding area are A number or a similar value is calculated, and when the value exceeds a certain threshold, the normalizer pixel is set as a black pixel.

(2) Normalization by opening and pulling Find a corresponding point on the pre-normalization image for each pixel on the normalized image, and use the value of the corresponding point as the value of the normalized pixel.

■Nonlinear normalization There are various methods. - The state of the distribution of character images within the film is extracted, and normalization is performed so that dense areas are widened and sparse areas are compressed.

[Problem to be solved by the invention]

However, the normalization method described above has a problem in that images are likely to be "blurred" or "fainted," and the regularization method using the threshold value and the nonlinear normalization method described above also have a problem in that the processing time is long.

In the case of character recognition, the number and position of holes in a single character image are often used as character features, so
``fading or blurring of thin line segments'' leads to the disappearance of holes, which has a large effect on the recognition results.

Therefore, it is an object of the present invention to provide an image normalization method that is suitable for character image normalization processing in character recognition, which is less likely to cause "smashing" or "blurring" that would lead to the disappearance of such holes, and which does not require much processing time. There is a particular thing.

[Means to solve the problem]

In order to achieve the above object, the image normalization method of the present invention determines a corresponding point on the normalized image for each pixel on the normalized image based on the ratio of the image sizes before and after normalization,
When the corresponding point on the pre-normalization image for a pixel of interest on the image after normalization is a black pixel, the pixel of interest is considered a black pixel, and the corresponding point on the image before normalization for each of the neighboring pixels of interest is a black pixel. When all pixels are white and there is a black pixel between the corresponding points, one pixel among the adjacent pixels of interest is considered a black pixel, and pixels on the normalized image that do not satisfy any of the above conditions are white. This is a pixel.

[For production]

To simplify the explanation, we will consider it in one dimension.

The size of the normalized double image is R51z6, and the size of the image before normalization is ○5ize. Then, the corresponding point on the pre-normalization image for each pixel on the post-normalization image is determined by the following equation.

T [X co = 0sizeX X/ R51ze...
Equation (1) However, or is the coordinate of the normalized pixel, and T[X is the corresponding point.

Then, in the case of R5ize (O5ize), the color (white, black) of the pixel after normalization is determined according to the following conditions.

a) Corresponding point of the pixel (X) of interest on the image after normalization (
If T [X] ) is a black pixel, the pixel of interest (X) is set as a black pixel.

b) Corresponding points between the pixel of interest (X) and the pixel next to it (X+1) on the normalized image (T[X] and T[X+L])
If both of these pixels are white pixels and there is a black pixel between these corresponding points, the pixel of interest (X) or the adjacent pixel (X+1) is determined to be a black pixel.

C) A pixel (X) for which neither the above condition a) nor the above condition b) is satisfied is a white pixel.

In the case of R5ize≧0size, when the corresponding point obtained by equation (1) is a white pixel, the normalizer pixel is set as a white pixel, and when the corresponding point is a black pixel, the normalizer pixel is set as a black pixel.

When applied to a practical two-dimensional image, equation (1) is applied to each direction to find corresponding points, and step a.

Rules b and c may be applied to each direction.

According to such a normalization method, portions that were separated in the image before normalization remain separate images even after normalization, and thin line segments on the image before normalization are preserved even after normalization. Therefore, if applied to the normalization of character images for character recognition, it is possible to prevent features such as holes from disappearing due to ``blurring'' or ``blurring''. Furthermore, since the algorithm is simple, the processing time can be reduced compared to a normalization method using a threshold value or a nonlinear normalization method.

〔Example〕

FIG. 1 is a schematic block diagram of a normalization device.

The pre-normalized image is stored in the image memory 10. The corresponding point determining unit 12 determines the coordinates of the corresponding points from the vertical and horizontal sizes of the pre-normalized image stored in the 1-image memory 0 and the vertical and horizontal sizes of the normalized image specified in advance, and notifies the converting unit 14 of the coordinates. 1
6 is an image memory for storing the normalized image.

The conversion unit 14 reads out pixel information from the image memory 10 regarding the corresponding points of the normalized pixels according to the corresponding point coordinate information notified by the corresponding point determination unit 12, and determines the color (white, black) of the normalized pixels. Then, the information “OI+” or “Engine” is written in the image memory 16.

A specific example of normalization will be explained with reference to FIG. In FIG. 2, 20 is a 120×120 pixel image before normalization, 22 is a 9×9 pixel portion of the pre-normalization image 20, and 24 is a 30×30 pixel portion.
Pixel normalized image, 26 is 3×3 of normalized image 24
This is the pixel part.

When the corresponding points are determined using the above equation (1), the result is as shown in the numerical table.

No. 1 schedule For example, the color of the normalizer pixel (0, O) is determined.

The corresponding point ■ is a white pixel. The corresponding point ■ of the adjacent pixel (0, 1) is a white pixel, and there is no black pixel between the corresponding points ■■. Further, the corresponding point (■) of the adjacent pixel (1, 0) in the other direction is a white pixel, and there is no black pixel between the corresponding points (■). That is, since it does not match either rule a or b, the normalized pixel (0°O) is determined to be a white pixel according to rule C.

The corresponding point ■ of the normalizer pixel (0, l) is a white pixel. The corresponding point ■ of the adjacent pixel (0, 2) is also a white pixel, and the corresponding point ■
There are no black pixels between ■. The corresponding point ■ of the adjacent pixel (1, L) in the other direction is a black pixel.

That is, since it does not match either rule a or b, the normalizer pixel (0, ↓) is determined to be a white pixel according to rule C.

The corresponding point ■ of the normalizer pixel (1, O) is a white pixel, but
Since the corresponding point ■ of the adjacent pixel (1, 1) is a black pixel, ■
and ■ does not match either rule a or b. However, the corresponding point ■ of the adjacent pixel (2, O) is a white pixel, and there is a black pixel between the corresponding points Decide on pixels.

However, the adjacent pixel may be a black pixel.

Since the corresponding point (2) of the normalizer pixel (L, L) is a black pixel, the normalizer pixel (1°1) is set as a black pixel according to the rule a.

[Effects of the Invention] As is clear from the above explanation, the normalization method of the present invention is as follows:
It is less likely to cause "blurring" or "blurring" that causes holes, which are important features of character images, to disappear, making it suitable for normalizing character images in character recognition devices, and the normalization algorithm is simple and fast. normalization processing becomes possible.

[Brief explanation of drawings]

It is a figure showing images before and after. 10... Image memory before normalization, 12... Corresponding point determination section, 14... Conversion section, 16.
...Normalized double image memory, 20.. Image before normalization, 24.. Normalization machine image. Figure 2

Claims (1)

[Claims] (1) A normalization method for image size, in which a corresponding point on the normalized image for each pixel on the normalized image is determined from the ratio of the image sizes before and after normalization, and normalization is performed. When the corresponding point on the pre-normalization image to the pixel of interest on the subsequent image is a black pixel, the pixel of interest is considered a black pixel, and the corresponding points on the pre-normalization image for each of the neighboring pixels of interest are all white. pixel and there is a black pixel between the corresponding points, one pixel among the adjacent pixels of interest is designated as a black pixel,
An image normalization method characterized in that pixels on a normalized image that do not satisfy any of the above conditions are treated as white pixels.