KR101929669B1 - The method and apparatus for analyzing an image using an entropy - Google Patents

Abstract

According to an embodiment of the present invention, a method for analyzing an image by using entropy comprises the following steps of: obtaining an image with respect to a product; performing Otsu binarization modified to maximize entropy with respect to the obtained image; distinguishing a background and a foreground from the binarized image; and determining whether a defect exists on the image based on the distinguished foreground. The Otsu binarization modified to maximize entropy is to binarize an image so as to maximize image entropy and a between-class variance of a threshold image at the same time.

Description

METHOD AND APPARATUS FOR ANALYZING AN IMAGE USING AN ENTROPY FIELD OF THE INVENTION [0001]

The present invention relates to a method and apparatus for analyzing an image using entropy, and more particularly, to applying an entropy weight to an Otsu method to simultaneously maximize the variance between an image entropy and a class of a thresholded image, And more particularly to an image analysis method and apparatus for displaying as many defective areas as possible in an image.

As a result of the availability of advanced technology and the rapid growth of industrial productivity, a tremendous amount of products are produced daily. This large-scale production makes manual fault detection impossible because of the huge amount of products that need to be inspected in a short period of time. Therefore, recent research on defect detection system based on equipment has been attracting attention. Two basic inspection methods for defect detection are Destructive Test and NonDestructive Test. In destructive testing, the product must be dismantled to check the physical properties such as strength, durability, and strength of the material or product. In addition, a destructive inspection can be used to reveal hidden defects that can not be determined by design or otherwise detected by design. Unlike this destructive test, NDT is a technique used to examine the properties of a product or to detect defects without altering or damaging the original properties of the product or the like. NDT is a good way to save time and money on defect detection, since the product can function normally after inspection. Typical NDT methods include ultrasonic nondestructive testing, thermal imaging NDT, visual NDT, liquid penetration NDT, and magnetic particle testing.

Of the general NDT methods, the naked eye nondestructive inspection is the fastest process and is preferred because it does not require complex equipment. The method generates visualization data of a material or a product using a camera, a scanner, and a sensor, and performs a process of detecting a defect from visualization data using a computer vision algorithm or an image processing technique. Such conventional visual nondestructive inspection has been relatively successfully applied for product defect detection. Different algorithms can be applied to inspect different types of products, such as textiles, paper, foil or steel, but since the majority of these algorithms are rather complicated, it takes a lot of time to complete the product defect inspection process. Therefore, a fast and automatic algorithm is needed which can produce satisfactory results for various products without manual manipulation by the user. One of the promising approaches for developing such a method is Automatic Thresholding.

Automatic thresholding is widely used not only in automated visual quality inspection systems, but also in many areas of computer vision, such as segmentation detection, edge detection, and motion analysis. These image analysis techniques require the process of rigorously selecting the optimal thresholds to divide the image into sets of backgrounds and objects of interest. That is, a gray scale image is converted into a k-ary image to separate the object from the background. Since automatic thresholding is a relatively simple and effective method, various application forms have been developed in image processing. That is, a large number of automatic thresholding related technologies have been developed. However, such conventional automatic thresholding techniques have exhibited variable efficiency (e.g., the efficiency is gradually decreased) in defect detection depending on the surface image characteristics (e.g., uniformity of pixel values) of the product, and are not stable.

1. Korean Registered Patent No. 10-1224164 (Jan. 13, 2014) 2. Korean Patent Registration No. 10-1612054 (Apr. 26, 2016)

The present invention has been devised as a countermeasure to the above-mentioned problem, and a new visual inspection algorithm which has improved the existing Otsu binarization is proposed.

Defect detection is one of the most important tasks in industrial quality control and is a difficult problem. Of the various visual inspection techniques, Automatic Thresholding is a method generally used for defect detection because the implementation into the system and the calculation of various parameters are relatively simple. In this conventional automatic thresholding principle, entropy We propose an automatic thresholding technique that dramatically improves Otsu's method by applying the Entropy Weighting Scheme.

As an embodiment of the present invention, a method and apparatus for analyzing an image using entropy can be provided.

A method for analyzing an image using entropy according to an embodiment of the present invention includes the steps of obtaining an image for a product, performing Otsu binarization so that entropy is maximized for the obtained image, Determining the presence or absence of a defect on the image based on the divided foreground and separating the background from the foreground in the extracted image and correcting the entropy to maximize the Ots binarization, May be to binarize the images so that they are maximized at the same time.

The modified Ots binarization is binarization in which image entropy is estimated by weighting by the following equation in the process of estimating the threshold value for maximizing the variance between classes for the image,

는 이미지의 엔트로피 함수, L은 그레이 레벨(gray level),

는 제 1 클래스 발생 확률,

는 제 1 클래스의 평균 그레이 값,

는 제 2 클래스 발생 확률,

는 제 2 클래스의 평균 그레의 값일 수 있다. k * is the threshold,

Is the entropy function of the image, L is the gray level,

Is the first class occurrence probability,

Is the average gray value of the first class,

The second class occurrence probability,

May be the value of the average gray of the second class.

The image of the product according to an exemplary embodiment of the present invention is a surface image of the product, and it is possible to detect a defective area in the surface image of the product through modified Ots binarization without receiving a separate parameter from the user.

An apparatus for analyzing an image using entropy according to an embodiment of the present invention includes an image obtaining unit for obtaining an image for a product, an image for performing Otsu binarization so that entropy is maximized for the obtained image, And an image analysis unit for discriminating the background and the foreground from the binarized image and determining the presence or absence of a defect on the image based on the divided foreground. The Otsu binarization modified to maximize the entropy includes an image entropy and a thresholded image May be to binarize the images so that the inter-class variance of the images is maximized at the same time.

The modified Ots binarization may be a binarization in which the image entropy is estimated to be weighted in the process of estimating the threshold to maximize the inter-class variance for the image.

Also, the image for the product is the surface image of the product, and the defect area can be detected from the surface image of the product through the modified Ots binarization without receiving any additional parameters from the user.

Meanwhile, as an embodiment of the present invention, a computer-readable recording medium on which a program for causing the computer to execute the above-described method may be provided.

In the method and apparatus for analyzing an image using entropy according to an embodiment of the present invention, entropy theory can be used to detect the presence or absence of a defect on a surface of a product without receiving a separate parameter from a user. That is, the user does not need to manually adjust the parameters and the like for threshold determination.

In addition, the method of analyzing an image using entropy according to an embodiment of the present invention can be performed completely automatically after an image of a product is obtained, and it is possible to detect a defective area that is extremely small as compared with the surface area of the product have.

In addition, the method of analyzing an image using entropy according to an embodiment of the present invention is advantageous in that since the calculation of threshold values and the like is simple, an analysis process for an image can be processed in real time, System and the like.

Figure 1 shows the efficiency of the valley-emphasis method versus the conventional Otsu's method for defect detection.
Figure 2 shows the histogram and the selected threshold of the image of Figure 1;
3 shows the results of applying the conventional Otsu's method and the like.
Figure 4 shows an example image of a wafer surface and a histogram of such an image.
FIG. 5 shows a flowchart of a method of analyzing an image using entropy according to an embodiment of the present invention.
6 is a block diagram of an apparatus for analyzing an image using entropy according to an embodiment of the present invention.
Figure 7 shows the effect (performance) of the method according to an embodiment of the present invention in defect detection in the product surface image.
Figure 8 shows the results of detecting defects using the method and the conventional method according to an embodiment of the present invention.
Fig. 9 shows an enlarged portion of a region indicated by a white rectangle in Fig. 8 (a).

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings, which will be readily apparent to those skilled in the art. The present invention may, however, be embodied in many different forms and should not be construed as limited to the embodiments set forth herein. In order to clearly illustrate the present invention, parts not related to the description are omitted, and similar parts are denoted by like reference characters throughout the specification.

The terms used in this specification will be briefly described and the present invention will be described in detail.

While the present invention has been described in connection with what is presently considered to be the most practical and preferred embodiment, it is to be understood that the invention is not limited to the disclosed embodiments. Also, in certain cases, there may be a term selected arbitrarily by the applicant, in which case the meaning thereof will be described in detail in the description of the corresponding invention. Therefore, the term used in the present invention should be defined based on the meaning of the term, not on the name of a simple term, but on the entire contents of the present invention.

When an element is referred to as "including" an element throughout the specification, it is to be understood that the element may include other elements as well, without departing from the spirit or scope of the present invention. Also, the terms "part," " module, "and the like described in the specification mean units for processing at least one function or operation, which may be implemented in hardware or software or a combination of hardware and software . In addition, when a part is referred to as being "connected" to another part throughout the specification, it includes not only "directly connected" but also "connected with other part in between".

Otsu's method, a state-of-the-art automatic thresholding technique, is the basis of many automatic defect detection schemes. This method relates to determining an optimal threshold that maximizes the difference between foreground and background classes. Otsu's method, an automatic clustering-based image thresholding method, can be understood as follows.

Let I = f (x, y) be a digital image, where f is an intensity function. F (x, y) has a gray value in the (x, y) region in the [0, L-1] region. Here, L represents the number of gray levels. In addition, assuming that the number of pixels of the gray values i and n _i, as N the total number of pixels in I In this case, the histogram can be considered in the following distribution of Equation (1), such as.

[Equation 1]

At this time, the average gray value I can be expressed by the following equation (2).

&Quot; (2) "

As a result of the thresholding, the pixels of I are given by C ₁ = {(x, y) | 0 ≤ f (x, y) ≤ k} and C ₂ = {(x, y) | k + 1 ≤ f (x, y) ≤ L - 1}, where k represents the selected threshold. In general, C ₁ may be foreground and C ₂ may be background. The class occurrence probability and the average gray value of each class can be estimated (computed) using Equation (3) and Equation (4) as follows.

&Quot; (3) "

&Quot; (4) "

In Otsu's method, the performance of the resulting threshold can be measured taking into account the difference between foreground and background. Using this criterion, the optimal threshold value k * Must be the maximum between the class variance values (see Equation 5 below).

&Quot; (5) "

In addition, another equivalent equation for calculating the optimal threshold value can be expressed as Equation (6) below. Equation (6) can be expressed more simply than Equation (5).

&Quot; (6) "

Otsu's method can provide appropriate results when the objects in the image are large and sufficiently different from the background. More specifically, Otsu's method may be effective in images where the histogram has a bimodal distribution or a multimodal distribution. However, since the histogram of the image obtained from the visual inspection system is smaller than the size of the image, the defect region appears to be a single-pole distribution. This feature of the image prevents Otsu's method from determining the optimal threshold value, and thus Otsu's method is not suitable for detecting defects in such images.

As described above, various improvements have been proposed in connection with defect detection using Otsu's method. The basic idea is to add a weight W to Equation 6, an objective function, to adjust the output threshold. (See Equation 7)

&Quot; (7) "

The earliest weighting method for Otsu's method is the Valley-emphasis method. (See Equation 8)

&Quot; (8) "

Here, the weight W is 1 - p _k . The core of the Valley-emphasis method is to complement the probability of occurrence (1 - p _k ). The smaller the value of p _k (ie, the lower the probability of appearance of gray value k), the greater the weight. These weights are named "valley-emphasis" because the output threshold is always at the valley or bottom boundary of the histogram. Figure 1 shows the efficiency of the valley-emphasis method versus Otsu's method for defect detection. The selected threshold value may be indicated by a vertical red line as in FIG.

Following the Valley-emphasis method, a weighting method for Otsu's method, which is also used for the probability of occurrence, has been proposed. There is also a weighted method based on Fisher's Linear Discriminant Analysis (LDA). Figure 3 shows the threshold of the result of implementing the methods described above.

As in FIG. 3, although the methods described above significantly improve the accuracy of defect detection, the performance of these methods is not stable. Moreover, some methods require the user to select appropriate parameters in order to obtain acceptable results.

In summary, Otsu's method is effective in thresholding the histogram of a bimodal distribution or a multimodal distribution. The Otsu method is not applicable when the histogram is a unimodal distribution or close to a singleton distribution. Thus, this conventional OTS method can provide satisfactory results for thresholding a general real image, but is not applicable to a visual inspection system due to the characteristics of the input image (ex. Histogram is a monoblock distribution, etc.). Since the histogram of the actual image is usually larger than the image size, it has a multivariate distribution. By using this characteristic, the optimal threshold value can be easily determined by Otsu's method. However, the histogram of the image obtained from the visual inspection system usually has a singular distribution due to the uniformity of the size of the defective area and the luminance, which hinders the identification of the desired threshold value using Otsu's method. Various remedies have been proposed to overcome the disadvantages of Otsu's method in defect detection, but these remedies have not yielded significant benefits in some cases, especially when detecting small size defects. In addition, some methods require parameters that must be manually assigned at the user's discretion. For this reason, the conventional methods are disadvantageous in that they can not reliably guarantee the reliability of defect detection when applied to other inspection systems.

The method and apparatus for analyzing an image using entropy according to an embodiment of the present invention is advantageous in that entropy theory is used and parameter of user input is not separately required. In addition, the image analysis process is fully automatic, and can detect areas of defects that are extremely small compared to the surface area of the product, quickly and accurately.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, the present invention will be described in detail with reference to the accompanying drawings.

Image entropy is a statistical measure of the randomness that can be used to characterize an image. Low entropy images contain a series of pixels with little or no information and typically with similar gray values. The entropy of an image where all pixels have the same gray value is zero. In the conventional entropy-based weighting method, given the gray value k, the posteriori entropy of the image can be defined as:

&Quot; (9) "

&Quot; (10) "

을 최대로 한다면, 다음과 같은 자명한 결과를 획득할 수 있다. (수학식 11 참조)if

, The following obvious results can be obtained. (See Equation 11)

&Quot; (11) "

This problem can be solved by transforming Equation (11) into an objective function.

&Quot; (12) "

&Quot; (13) "

A gray value k with a maximum value of Ψ (k) can be used as a threshold value because it maximizes the distinction between the object on the image and the background. However, if it is uniform with the distribution of the object and the background, Equation (12) can be expressed as follows.

&Quot; (14) "

And, when k = 1/2 (L - 1), the maximum value can be reached. This property can be a disadvantage when applied to the surface defect detection of a product. As shown in Fig. 4, there is a problem that this characteristic reduces the adaptability of the defect detection algorithm because the distribution of the product surface image is generally uniform. In order to overcome this problem, the present invention combines the entropy objective function ψ (k) with Otsu's method. That is, in Equation (7), a new objective function is generated by replacing the weight W with the objective function? (K).

&Quot; (15) "

Using Equation (15), k can be obtained as an optimum value for selection of a threshold value. The basic concept of the weighting method applied to the present invention for Otsu's method is to simultaneously maximize the image entropy and the variance between the classes of the thresholded image. This is to show as many defective areas as possible in the visual inspection. Since the defective area is extremely small and random compared to the surface area of the product, the threshold for maximizing the image entropy can also maximize the number of pixels exposed in the defective area.

A method of analyzing an image using entropy according to an exemplary embodiment of the present invention includes the steps of acquiring an image of a product (S100), performing Otsu binarization so that entropy is maximized for the acquired image A step S300 of discriminating a background and a foreground in the binarized image and a step S400 of determining whether or not there is a defect on the image based on the divided foreground, May be to binarize the image so that the variance between the image entropy and the class of the thresholded image is maximized simultaneously.

The modified Ots binarization is binarization in which image entropy is estimated by weighting by the following equation in the process of estimating the threshold value for maximizing the variance between classes for the image,

는 이미지의 엔트로피 함수, L은 그레이 레벨(gray level),

는 제 1 클래스 발생 확률,

는 제 1 클래스의 평균 그레이 값,

는 제 2 클래스 발생 확률,

는 제 2 클래스의 평균 그레의 값일 수 있다. k * is the threshold,

Is the entropy function of the image, L is the gray level,

Is the first class occurrence probability,

Is the average gray value of the first class,

The second class occurrence probability,

May be the value of the average gray of the second class.

The image of the product according to an exemplary embodiment of the present invention is a surface image of the product, and it is possible to detect a defective area in the surface image of the product through modified Ots binarization without receiving a separate parameter from the user.

An apparatus 1000 for analyzing an image using entropy according to an exemplary embodiment of the present invention includes an image obtaining unit 100 for obtaining an image of a product, an Otsu And an image analyzer 300 for separating the background and the foreground from the binarized image and determining the presence or absence of a defect on the image based on the divided foreground, The modified Otsu binarization may be to binarize the image so that the variance between the image entropy and the class of the thresholded image is maximized at the same time. The image obtaining unit 100 may include a camera unit 110 and an illumination unit 120 according to an exemplary embodiment of the present invention.

The modified Ots binarization may be a binarization in which the image entropy is estimated to be weighted in the process of estimating the threshold to maximize the inter-class variance for the image.

Also, the image for the product is the surface image of the product, and the defect area can be detected from the surface image of the product through the modified Ots binarization without receiving any additional parameters from the user.

The effect (performance) of the method according to an embodiment of the present invention in defect detection in the product surface image was confirmed as in FIG. It has been confirmed that the method according to one embodiment of the present invention exhibits excellent performance as compared with FIG. 1 and FIG. It has also been found that the method according to one embodiment of the present invention exhibits stable and reliable performance when applied to various types of product surface patterns.

Also, we compared Otsu's method and 4 weighting methods for a single image as shown in Fig. For how to specify the parameters, the recommended parameters are used.

Figure 8 (a) shows a sample test image with several defective areas to compare several methods, and Figures 8 (b) to (g) show the thresholding results obtained from the six methods described above . It can be seen that the output threshold of the method according to an embodiment of the present invention shows another defective area except for a clearly visible area, but in other methods it fails to detect or only detects a clear area.

Fig. 9 shows an enlarged portion of a region indicated by a white rectangle in Fig. 8 (a). The Otsu's method and the Gaussian-weight valley-emphasis method can not detect the defective area of the sample test image and are omitted in FIG.

In addition, for the further evaluation, ten identical-sized wafer images with a resolution of 3000 x 1500 pixels and a gray level of L = 256 were used. Each image had a very small defective area compared to the image size. The results are shown in Tables 1 and 2. As shown in Tables 1 and 2, the defective area detected by the method according to an embodiment of the present invention is larger than the area detected by other methods. It can also be seen that the output threshold value of the method according to an embodiment of the present invention has the lowest deviation while the output threshold value according to other methods has a significant deviation.

[Table 1]

[Table 2]

The contents of the above-described method can be applied in connection with the apparatus according to an embodiment of the present invention. Therefore, the description of the same contents as those of the above-described method with respect to the apparatus is omitted.

On the other hand, the above-described method can be implemented in a general-purpose digital computer that can be created as a program that can be executed in a computer and operates the program using a computer-readable medium. Further, the structure of the data used in the above-described method can be recorded on a computer-readable medium through various means. Recording media that record executable computer programs or code for carrying out the various methods of the present invention should not be understood to include transient objects such as carrier waves or signals. The computer-readable medium may comprise a storage medium such as a magnetic storage medium (e.g., ROM, floppy disk, hard disk, etc.), optical readable medium (e.g., CD ROM, DVD, etc.).

It will be understood by those skilled in the art that the foregoing description of the present invention is for illustrative purposes only and that those of ordinary skill in the art can readily understand that various changes and modifications may be made without departing from the spirit or essential characteristics of the present invention. will be. It is therefore to be understood that the above-described embodiments are illustrative in all aspects and not restrictive. For example, each component described as a single entity may be distributed and implemented, and components described as being distributed may also be implemented in a combined form.

The scope of the present invention is defined by the appended claims rather than the detailed description and all changes or modifications derived from the meaning and scope of the claims and their equivalents are to be construed as being included within the scope of the present invention do.

100: image obtaining unit 110: camera unit
120:
200: image conversion unit 300: image analysis unit
1000: Image analysis using entropy

Claims (7)

A method for analyzing an image using entropy,
Obtaining an image for the product;
Performing modified Otsu binarization to maximize entropy for the obtained image;
Separating the background and the foreground in the binarized image; And
Determining whether or not there is a defect on the image based on the divided foreground,
The modified Ots binarization to maximize the entropy is to binarize the image so that the image entropy and the between-class variance of the images classified as a result of the thresholding are maximized simultaneously,
Wherein the class of the image represents a group of pixels classified on the basis of an estimated threshold value in the modified Ots binarization process.
The method according to claim 1,
Wherein the modified Ots binarization is a binarization in which the image entropy is weighted and estimated by the following equation in a process of estimating a threshold value such that a variance between classes for the image is maximized,

k * is the threshold,

Is an entropy function of the image, L is a gray level,

Is the first class occurrence probability,

Is the average gray value of the first class,

The second class occurrence probability,

Is a method of analyzing an image using entropy, which is a value of the mean gray of the second class.
The method according to claim 1,
Wherein the image for the product is a surface image of the product,
Wherein the defect region can be detected in the surface image of the product through the modified Ots binarization without receiving a separate parameter from the user.
An apparatus for analyzing an image using entropy,
An image obtaining unit for obtaining an image of a product;
An image transforming unit for performing Otsu binarization so as to maximize entropy with respect to the obtained image; And
And an image analyzer for separating the background and the foreground from the binarized image and determining the presence or absence of a defect on the image based on the divided foreground,
The modified Ots binarization to maximize the entropy is to binarize the image so that the image entropy and the between-class variance of the images classified as a result of the thresholding are maximized simultaneously,
Wherein the class of the image represents a group of pixels classified on the basis of an estimated threshold value in the modified Ots2 binarization process.
5. The method of claim 4,
Wherein the modified Ots binarization is a binarization in which the image entropy is weighted and estimated by the following equation in a process of estimating a threshold value such that a variance between classes for the image is maximized,

k * is the threshold,

Is an entropy function of the image, L is a gray level,

Is the first class occurrence probability,

Is the average gray value of the first class,

The second class occurrence probability,

Is an entropy that is a value of the mean gray of the second class.
5. The method of claim 4,
Wherein the image for the product is a surface image of the product,
Wherein the defect region can be detected in the surface image of the product through the modified Ots binarization without receiving a separate parameter from the user.
A computer-readable recording medium on which a program for implementing the method of any one of claims 1 to 3 is recorded.

Images