JPH0337783A - Image normalizing system - Google Patents

Abstract

PURPOSE:To handle in a lump images whose size, etc., are different variously without generating 'blur' and collapse of an image by executing a thinning processing of a segment as a pre-processing of an OR (logical sum) normalization processing. CONSTITUTION:In a normalization processing part 5, based on information of a size, etc., of a character delivered by a thinning processing direction/frequency control part 6, the direction for applying the thinning processing and the number of times are determined and designated to a thinning processing part 7, and by the thinning processing part 7, the thinning processing of a segment from the designated direction is executed by a designated frequency portion with respect to a character image in a character image buffer 4. The character image after the thinning processing is left in the character image buffer 4. When this thinning processing is ended, an OR processing part 8 executes an OR normalization processing to the character image which is read in from the character image buffer 4, and data of a normalized character image which is generated is stored in a normalized image buffer 9 in an image memory 3. With respect to this image, a character recognizing part 10 executes a character recognition processing. In such a way, a normalized image being free from a collapse of a space and disappearance or 'blur' of a segment can be obtained, and various images can be processed in a lump.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to an image normalization method for normalizing the size of one image.

[Conventional technology]

In a device or system that reads and recognizes images of characters of various sizes, it is necessary to normalize the size of the input image as preprocessing for character recognition.

Basic techniques for normalizing such images include 5-decimation normalization and OR normalization.

As shown in FIG. 4, thinning normalization is a method of extracting pixels at regular intervals and discarding other pixels to generate a normalized image smaller in size than the input image. As shown in FIG. 5, OR normalization is a method of generating a normalized image smaller in size than the input image by calculating the logical sum (OR) of pixels within a certain interval.

However, when such a method is applied to the normalization of character images, thinning normalization tends to cause thin line segments to disappear or become "fainted," while OR normalization tends to cause thin blank spaces to collapse. This is known to be the cause of erroneous recognition of characters. Examples of such collapse or "fading" are shown in FIGS. 6 and 7.

Considering these problems, as a preprocessing for thinning normalization, we detect thin parts of line segments and add at least one
There is a method to prevent line segments from disappearing or 'fading' by adding pixels (Japanese Patent Laid-Open No. 60-126780), and an image in which the input image is OR-normalized as it is, and an image in which the input image is horizontally and vertically A method (Japanese Unexamined Patent Application Publication No. 110086/1986) has been devised to prevent blank spaces within characters from being crushed by logically multiplying images that have been OR-normalized by shifting one pixel in each direction.

[Problem to be solved by the invention]

However, when either method is applied to an image larger than twice the normalized size, the processing becomes complicated and a sufficient effect cannot be obtained.

Therefore, an object of the present invention is to provide an image normalization method that can handle images of various sizes etc. all at once without causing "blurring" or blurring of the images.

[Means to solve the problem]

The feature of the image normalization method of the present invention is that line segment thinning processing is performed on the input image as a preprocessing process for generating a normalized image using a logical sum operation (OR normalization processing), which is a basic normalization method. It is to do.

[For production]

By thinning the line segment before the OR normalization process, it is possible to compensate for the weakness of the OR normalization, that is, the tendency for thin blank spaces to collapse. Also, OR normalization is essentially 1
Unlike thinning normalization, this method has the advantage that thin line segments are less likely to become blurred or disappear. Therefore, a normalized image without image distortion or blurring can be obtained.

In addition, by simply optimizing the effect of the thinning process by changing the number of times and direction of thinning processing depending on the size of the input image, etc., it is possible to normalize images without distortion or "blurring" for various images of different sizes, etc. You can get images.

〔Example〕

An embodiment of the present invention will be described below using the character recognition system shown in FIG.

From the original image data input from the original image input unit (file, scanner, etc.) 1, the character extraction unit 2 extracts image data for one character and stores it in the character image buffer 4 in the 1-image memory 3. Information such as the size of is passed to the normalization processing unit 5. This normalization processing unit 5 is a part directly related to the present invention.

In the normalization processing section 5, the narrowing processing direction/number of times control section 6
1, which determines the direction and number of times to apply thinning processing based on the passed information such as the size of the characters, and specifies it to the thinning processing section 7.
Next, the narrowing processing unit 7 executes line segment thinning processing from the specified direction on the character image in the single character image buffer 4 for the specified number of times.

The character image after the narrowing process remains in the character image buffer 4. When this narrowing process is completed, the OR processing unit 8 performs OR normalization processing on the character image read from the character image buffer 4, and the data of the generated normalized character image is transferred to the normalized image buffer 9 in the image memory 3. Store in. The character recognition unit 10 executes character recognition processing on this image.

The narrowing process will be explained with reference to FIGS. 2 and 3. There are four possible narrowing directions: top to bottom, bottom to top, left to right, and right to left.

FIG. 2 is an explanatory diagram of narrowing processing from top to bottom.

This example looks at the pixels from top to bottom, and if two or more black pixels overlap in that direction, replace the topmost black pixel with one white pixel (delete one black pixel). So,
The top black pixel in the ↓ mark part is erased, but since there is only one black pixel in the 0 mark part, it is not erased.

FIG. 3 is an explanatory diagram of narrowing processing from bottom to top.

In this case, in the area where two or more black pixels overlap from bottom to top (the area marked ↑), one black pixel at the bottom is erased, but in the area marked 0, there is only one black pixel. So I will leave it without erasing it.

Through such processing, a line segment in the horizontal direction can be narrowed without causing any break in the line segment.

The thinning process in the left and right direction is similar, and it is possible to narrow the line segment in the vertical direction without causing any break in the line segment.

The direction of thinning processing is determined depending on the characteristics of the character image6. For example, if the scanner used to read the original image has a characteristic that easily erases horizontal line segments of characters, or if the target is a font with thick horizontal line segments. In some cases, only the vertical direction is selected. On the other hand, in the case of a scanner in which the vertical lines of characters are easily crushed, or in the case of a font with thick vertical lines, thinning processing only in the horizontal direction is selected. Naturally, the up, down, left, and right directions may be selected.

By appropriately determining the direction of thinning processing in this manner, it is possible to effectively prevent blank spaces from disappearing or "fading" in various character images without causing lines to disappear or "fading".

Further, the number of times of narrowing processing is determined according to the image size, for example, by the following equation.

[Number of narrowing processes] = [Vertical image size]/[Normalized size] By controlling the number of narrowing processes in this way, images of various sizes can be appropriately narrowed down.

Specifically, for a character image whose normalized size is twice as large in the vertical direction with small gaps between line segments in the horizontal direction, for example, thinning processing from top to bottom and thinning processing from bottom to top are performed. Apply once each alternately. The reason why the thinning processing is applied alternately from opposite directions in this way is to disperse the influence of image distortion caused by the narrowing processing.

Since the narrowing process as described above is performed as preprocessing, the normalized character image generated by the OR processing unit 8 includes O
The weak point of R normalization is that the white is less likely to collapse, and since it uses the OR normalization method, thin line segments are less likely to disappear or become "fainted" as would be the case with thinning normalization. Therefore, it is possible to reduce erroneous recognition of characters due to blurring or "fading".

〔Effect of the invention〕

As explained above, according to the present invention, line segment narrowing processing is performed as a preprocessing of OR normalization processing, so that the weaknesses of OR normalization can be compensated for without compromising its strengths, and the It is possible to obtain a normalized image without any loss or "fading", and it is also possible to process various images at once by simply controlling the direction and number of thinning processes depending on the nature and size of the target image.

[Brief explanation of drawings]

FIG. 1 is a block diagram of a character recognition system according to the present invention,
Figure 2 is an explanatory diagram of thinning processing from top to bottom, Figure 3 is an illustration of narrowing processing from bottom to top, Figure 4 is an illustration of thinning normalization, and Figure 5 is an illustration of OR. Explanatory diagram of normalization, 6th
The figure shows an example of thinning normalization, and FIG. 7 shows an example of OR normalization. 1...Original image input section, 2...Character cutting section, 3
...Image memory, 4...Character image buffer, 5.
...Normalization processing unit, 6...Narrowing processing direction/number control unit, 7...Narrowing processing unit, 8...OR processing unit, 9...Normalized image buffer, 10...Character recognition Department. Figure ■ Figure 4 Figure 6 Figure 5 Figure 7

Claims (1)

[Claims] (1) An image normalization method characterized by performing line segment thinning processing on an input image, and normalizing the processed image using a logical sum operation.