KR19990005294A - Input Image Binarization Method of Pattern Recognition System - Google Patents

Abstract

The present invention relates to a binarization method for an input image used in a pattern recognition system. The method of the present invention obtains a global threshold value using a distribution of pixel values of the input image, and extracts a 3 * 3 mask from the input image. Calculates a local threshold value, compares the global threshold value with the local threshold value, and selects the global threshold value as the final threshold value if the global threshold value is greater than the local threshold value value; If the value is less than or equal to the local threshold, selecting the local threshold as a final threshold, and setting the central pixel value to one of binary values according to a comparison of the pixel value of the central pixel and the final threshold value. It includes. Therefore, by adopting global threshold determination and local threshold determination adaptively, it is possible to provide a more accurate binarized image required for pattern recognition.

Description

Input Image Binarization Method of Pattern Recognition System

The present invention relates to a method for binarizing an image for pattern recognition of an input image in a pattern recognition system. More particularly, the present invention relates to an input image binarization method using an adaptive global threshold and a local threshold in a pattern recognition system. It is about.

In the field of image processing and computer vision, computer-based automatic pattern recognition is one of the important problems. Pattern recognition is performed by comparing features of patterns present in the input image with features of a model, which is a pre-stored pattern. For pattern recognition, a process of converting an input image into a digitalized binary image signal is required. These binarization transformations are largely divided into a method using a global threshold and a method using a local threshold.

The global threshold determination method calculates an arbitrary threshold that minimizes the variance of two parts when dividing 256 gray levels into two parts using the distribution of the number of pixels for each gray level of the input image. The pixel of the input image is binarized to a value of 0 or 1 according to the calculated threshold.

The local threshold determination method uses N * N to calculate a threshold value for a mask center pixel using pixel values within an area designated by each mask, and zeros or zeros pixels in the input image according to the calculated threshold value. Binarization to a value of 1.

However, when binarizing an image for pattern recognition, since only one of the two methods described above is used, it is difficult to obtain accurate results in various images, particularly in an image having few parts representing objects such as documents. In other words, images of the same type as documents represent all information as line components, and the area occupied by them is about 10 to 20% of the entire image. There is no problem.

Therefore, an object of the present invention is to provide a method for generating a more accurate binarized image of an input image in which a part representing an object in the entire image in the pattern recognition system is small.

According to the present invention for achieving the above object, a binarization method for an input image used in a pattern recognition system, the method comprising: obtaining a global threshold value by using a pixel value distribution of the input image, and the input image 3 * 3 obtaining a local threshold using a mask; comparing the global threshold with the local threshold and selecting the global threshold as the final threshold if the global threshold is greater than the local threshold; Selecting the local threshold as a final threshold if the global threshold is less than or equal to the local threshold, and comparing the central pixel to a binary value according to a comparison of the pixel value of the central pixel and the final threshold. Characterized in that the step of setting to one value.

1 is a schematic block diagram of a pattern recognition system using a binarization method of the present invention;

2 is a flowchart illustrating a binarization method according to the present invention;

3A is a diagram illustrating a pattern recognized result of a binarized image using a local threshold determination method;

3B is a diagram illustrating a pattern recognized result of a binarized image using a global threshold determination method;

And 3c illustrates a pattern recognized result of a binarized image using the adaptive threshold determination method according to the present invention.

* Explanation of symbols for main parts of the drawings

10: scanner 20: binarization unit

30: segmentation unit 40: pattern contour extraction unit

50: classifier

The above and other objects and various advantages of the present invention will become more apparent from the preferred embodiments of the present invention described below with reference to the accompanying drawings.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings.

1, there is shown a schematic block diagram of a pattern recognition system using a binarization method according to the present invention.

First, the scanner 10 including an optical scanner or the like scans a document or an image on a document. The binarization unit 20 digitizes the binary image using the global threshold value and the local threshold value determination method as described in detail below from the scanned image.

The binary image digitized by the binarization unit 20 is partitioned by the segmentation unit 30 and provided to the pattern contour extracting unit 40. The pattern outline extractor 40 extracts a feature such as an outline of an image by using a contour extraction algorithm well known in the art. Features such as extracted contours are detected by straight lines and curves in the image using a Hough transform and then provided to the classifier 50. The classifier 50 determines the class to which the linear and curved patterns belong and finally outputs the recognized result.

FIG. 2 is a flowchart illustrating an image binarization method of the present invention performed by the binarization unit shown in FIG. 1.

First, in step 110, when a scanned image is input through a scanner, a local threshold is calculated for this scanned image. That is, the threshold value for the mask center pixel using the pixel value in the region designated by each mask using N * N, for example, a 3 * 3 mask for the input image. Calculate Local threshold ( ) Is defined as in Equation 1 below.

The variable used in Equation 1 is defined as in Equation 2 below.

In Equation 2, M is the number of all pixels in the image, P (n) is the histogram value of the pixel n, and Pi represents the pixel value (grey value) of the pixel i in the 3 * 3 mask.

Then, in step 120, a global threshold is determined. Global threshold determination is a method of finding an arbitrary threshold t that minimizes the variance of two parts when dividing 256 gray levels into two parts using a distribution of the number of pixels for each gray level of the input image. It can be expressed as Equation 3 below.

The variable used in Equation 1 is defined as follows.

: Histogram of pixel i

: The number of pixels whose pixel value is t or less

: The number of pixels whose pixel value is t or more

Variance of pixel values less than or equal to t

variance of pixel values above t

: Probability that the pixel value is less than or equal to t

: Probability that pixel value is greater than or equal to t

In the above equation (3), the value of t that minimizes sigma _w ² (t) is determined as the threshold value.

In step 130, the local threshold ( ) Is compared with the global threshold t to obtain a final threshold value Γ for a pixel disposed at the center of the 3 * 3 mask. In this case, the global threshold t is the local threshold ( If greater than, the global threshold t is chosen as the final threshold Γ (step 140), and the global threshold t is the local threshold ( Less than or equal to), the local threshold ( ) Is chosen as the final threshold Γ (step 150). This selection process uses a global threshold when the distribution of pixel values is large and a local threshold when the distribution of pixel values is small to remove noises such as points in the input image and white space inside the line. This prevents text adjacent to the line from sticking to it.

In step 160, the center pixel is set to one of zero or one of binary numbers by comparing the pixel value Pi of the center pixel of the 3 * 3 mask with the final threshold value Γ. At this time, if the pixel value Pi of the center pixel of the 3 * 3 mask is greater than the final threshold value Γ, the center pixel value Pi is binarized to binary 1 (step 170), and the center pixel of the 3 * 3 mask If the pixel value Pi has smaller than the final threshold value Γ, the center pixel value Pi is binarized to binary zero.

FIG. 3 is a diagram illustrating a result of binarizing a map image as an input image by different methods. FIG. 3A and FIG. 3B use local threshold determination and global threshold determination, and FIG. It is a result of binarizing a map image by adaptively using global threshold determination, thinning after removing isolated characters using the size of interconnected components in the binarized image. In FIG. 3A, there are small noises such as dots and many holes are formed inside the thick line at the lower right. In FIG. 3B, there are no small noises and no holes in the line component. Notice that the letter 6 in is attached to the adjacent line component. However, in FIG. 3C illustrating the result recognized using the present invention, it can be seen that the letter 6 in the upper left is removed separately from the line component and the points and holes inside the line are removed.

Therefore, in a map-like image in which all information is represented by a line component and the area occupied is about 10 to 20% of the entire image, the pattern by adaptively adopting global threshold determination and local threshold determination in accordance with the present invention. It is possible to provide more accurate binarized images for recognition.

Claims (2)

A binarization method for an input image used in a pattern recognition system, the method comprising: obtaining a global threshold value using a pixel value distribution of the input image, and obtaining a local threshold value using the 3 * 3 mask of the input image And comparing the global threshold value and the local threshold value to obtain a final threshold value for a pixel disposed at the center of the 3 * 3 mask, wherein the pixel value of the center pixel is compared with the final threshold value. And setting the center pixel value to one of binary digits. 2. The method of claim 1, wherein obtaining the final threshold selects the global threshold as a final threshold if the global threshold is greater than the local threshold, and wherein the global threshold is greater than the local threshold. If less than or equal to, selecting the local threshold as the final threshold.