JPS63219081A - Image binarization method - Google Patents

Abstract

PURPOSE:To remove the influence of the non-uniformity of the illumination and background in an image and the variety of the reflecting characteristic of an object by providing a threshold automatic determining function and an adaptive capacity to an image strength change depending on a place. CONSTITUTION:An input image 10, in which an illumination component and the reflecting characteristic of an object are non-uniform, is stored first through an image analyzing processing part 60 to an image file 90. next, for the stored image, a binarizing image to remove the non-conformity by the processing part 60 is calculated, and outputted to a conversational image displaying terminal 70. In this case, a filter to presume the noise of the illumination component, etc. is obtained and after a low frequency component is extracted and removed from the inside of the image by the filter, the image is divided into plural small blocks. A local threshold based on a discriminating analyzing method is calculated for respective blocks, the image is binarized (block threshold value determining processing) and displayed at a displaying terminal 70. Thus, the binarization to remove the influence of the non-uniformity of the illumination and background in the image and the variety of the reflecting characteristic of the object can be executed and the binarizing accuracy can be improved.

Description

[Detailed Description of the Invention] [Industrial Field of Application] The present invention relates to image analysis processing, particularly when the distribution of illumination components is uneven and there are objects with various reflection characteristics. The present invention relates to a binarization method suitable for binarizing images taken below.

[Prior art] The conventional image binarization method is as described in Nagao Makoto Iwahiro's Information Processing Science 21 "Pattern Recognition and Graphic Processing J, March 1983, Iwanami Shoten, pages 151 to 153. ,
There are (a) a 6-degree histogram valley detection method, (b) a discriminant analysis method, and (Q) an interactive method.

When generating a binary image consisting of a background and a target area from an observed grayscale image, variations in image intensity characteristics within one image, that is, non-uniformity of illumination and background, and diversity of reflection characteristics of the target are considered. There is a problem though.

Conventional methods have insufficient countermeasures against this problem.

[Problem that the invention seeks to solve]

The above-mentioned conventional technology does not take into consideration the two points of first order, and when a wide range of types of substances exist on the same image,
There was a problem with accuracy when converting into values.

(b) Illumination and background inconsistency - If the distribution of illumination in an image is non-uniform, the observed image intensity will differ depending on its position, even for objects with the same reflection characteristics. . Therefore, it is inappropriate to set a single threshold value for binarization for images in such a state. The same can be said for the heterogeneity of the background distribution.

(b) Diversity of reflection characteristics of objects: If there are multiple types of objects, each with different reflection characteristics, the image intensity in the observed image will vary greatly, as in (b).
It is inappropriate to set a single threshold.

SUMMARY OF THE INVENTION An object of the present invention is to provide a binarization formula that has an automatic threshold value determination function and an adaptability to location-dependent image intensity fluctuations in image binarization processing, which has the above-described problem.

[Means for solving problems]

The above purpose is (a) to remove the variation factor for non-uniformity of illumination and background by subtracting the low frequency component from the single measurement image. (Low-frequency components are obtained by passing the observed image through a low-pass filter.) (B) To account for the diversity of the reflection characteristics of the target, the image from which the low-frequency components have been removed is divided into multiple small blocks. This is achieved by calculating a local threshold value for each block based on a discriminant analysis method and binarizing it.

[Effect]

In the case of remote sensing images, the method (b) of removing non-uniformity in illumination and background by subtracting the low frequency components from the observed image is a method in which illumination such as sunlight Focusing on the fact that the spatial frequency of the distribution on the ground (and background image intensity distribution) is sufficiently low, low-frequency illumination or background components are removed.

(b) The image from which low frequency components have been removed is divided into multiple blocks, and a local threshold value is calculated for each block using a discriminant analysis method. Able to respond to fluctuations.

〔Example〕

An embodiment of the present invention will be described below with reference to FIGS. 1 to 3.

The embodiment is an image analysis processing system whose purpose is to obtain a binarized image from a grayscale satellite image.

Figure 2 shows an overview of this system. An input image 10 with non-uniform illumination components and reflection characteristics of an object is input and stored in an image file 90. A single image analysis processing section 60 calculates a binarized image from which the non-uniformity has been removed from the stored image and outputs it to the conversational image display terminal 7o.

When accurately estimating illumination components and the like in the image analysis processing unit 60, the image analysis processing system user 80 assumes the size of the object to be detected in the image and finds a filter for estimating noise such as the illumination components.

Details of the image analysis processing section 60 will be described below.

First, non-uniformity on the image caused by sunlight or the like is removed as follows.

In order to remove the low frequency components from the image, the input image f
(x, y) 10 is passed through a low-pass filter h (x, y) 20 having the characteristics shown in FIG. 3(a) to extract low frequency components.

The characteristics and determination method of the low-pass filter h Cxt y') are performed as follows.

If the output after image intensity difference calculation 30 is g (X, y), then g (x + y) = f (xt y) - f (
x+y)oh(x,y) (1) Mocking: In the convolution frequency range, G(ω8.ωy)=F(ω8.ωy)(1-IH(ω8
．． ω, )) (2) Here, G, F, and H are g, respectively.
This is a frequency domain representation of f and h.

From equation (2), G(ω8.ωy) is the input image F(ω8.ωy).
This is the result of applying a filter having a frequency characteristic expressed by (1-H, (ω8.ωy)) to ωy).

For example, if we set the following rectangular relationship as h(ω8.ωy), O; other than the above, in this case l H((,1,,ωy) l is shown in Figure 3(a)
This is the absolute value of the S inc function as shown in . However, in the figure, for simplicity, ω = 0 and a -dimensional representation is shown. On the other hand, the frequency characteristic represented by 1-IH(ω8.ωy)1 is shown in FIG. 3(b). In other words, this becomes the attenuation characteristic for the input image.

By the way, in order to effectively remove illumination components and background components from an input image, it is necessary to estimate the frequency characteristics of these components and determine the attenuation characteristics based on the frequency characteristics. However, it is generally difficult to accurately estimate illumination components and background components from an input image. Therefore, in this method, h(x, y) is determined by the following approximation. That is, the user of the image analysis system assumes the largest square that he wants to detect in the image, and sets the number of pixels on the negative side to the size N of one side of the domain of h(x, y). h(x,y) in this range
As the shape of , it is also possible to use a Gaussian function which has the advantage of not generating ripples in the frequency characteristics in other passbands than the rectangular function in equation (3). In the method for determining h(xty) described here, if the object has a side length of N pixels or less, it will not be removed together with the illumination component for the filter or the background component.

Next, in order to deal with the diversity of reflection characteristics, the image from which the low frequency components have been removed is divided into multiple small blocks as described above, and a local threshold value based on the discriminant analysis method is calculated for each block. Then perform image binarization. This is called block threshold value determination processing 40, and the result is output on a binarized image display 50.

The size of the block is related to the binarization accuracy.

If the blocks are made smaller, the binarization accuracy will improve, but the processing time will increase. Determined by the size of the object to be extracted. According to the embodiment, this method extracts only the outline of the object, and the noise in the image φ is also small. In addition, compared to processing the entire image with a single threshold, the binarized image produced by this method can express the structure of the object in more detail in areas with different image intensities, and has the effect of improving binarization accuracy. .

The present invention is applicable not only to image analysis systems for remote sensing but also to industrial inspections that require microscopic shape analysis, such as scanning electron microscopes (SEMs) or chromosome inspections in medicine.

〔Effect of the invention〕

According to the invention, illumination and background non-uniformity in the image;
Since binarization can be performed while removing the influence of the diversity of reflection characteristics of the object, there is an effect of improving the binarization accuracy.

[Brief explanation of drawings]

FIG. 1 is a flowchart showing the adaptive binarization processing procedure according to an embodiment of the present invention, FIG. 2 is a block diagram of the image analysis processing system, and FIG. 3 is a diagram of the low-pass filter and low-frequency component removal system. Zhou Diagram 3

Claims (1)

[Claims] 1. In an image analysis processing method consisting of binarization processing of pixel intensity and display processing of the binarized image, the low frequency components of the input image are extracted and the pixel intensity difference between the input image and the low frequency component image of the image is calculated. An image binarization method characterized in that a difference image after the calculation is divided into small regions, and a binarization process is performed for each of the divided small regions.