KR101677070B1 - System and Method for Automatically Detecting a Mura Defect using Morphological Image Processing and Labeling - Google Patents

Abstract

The present invention relates to a system and a method for automatically detecting a spot defect using morphological image processing and labeling,
A preprocessing unit for removing an illumination component from the image acquired by the image acquiring unit using morphological image processing; and a histogram acquisition unit for acquiring a histogram of the image from which the illumination component has been removed in the preprocessing unit A histogram smoothing unit for reconstructing an image by expanding to a whole gray level through histogram smoothing and a smoothing candidate generating unit for forming a smoothing candidate region by applying a Gabor filter to the image transformed by the image processing unit; It is possible to provide a system and a method for automatically detecting odd defect by using a video analyzing unit having a smear detecting unit for detecting a smear by using labeling in a smear candidate region formed in the candidate generating unit.

Description

Technical Field [0001] The present invention relates to a system and a method for automatically detecting blur defects using morphological image processing and labeling,

More particularly, the present invention relates to a system and method for automatically detecting blur defects using morphological image processing and labeling.

As the IT industry develops, the use of portable electronic products such as smart phones and tablet PCs, and LCD monitors, TVs, and other household appliances, It is actively proceeding.

If the display surface is contaminated with light or dirt, the light emitted from the backlight module is distorted. Therefore, when the display is blurred or worn by the user, the luminance non-uniformity -uniformity) is displayed, resulting in a drop in merchantability and inconvenience to the user, and there is no way to repair it. Therefore, quality inspection of the display surface is important in the manufacturing process. Therefore, it is necessary to confirm the mura of the display surface before shipment.

At this time, the unevenness may be caused by dust and foreign matter during the manufacturing process, scratches caused by the carelessness of the inspector, fingerprints, and the like. Until now, these stains have been directly inspected by the naked eye to determine the stain of the display film. However, when the stain is visually inspected, there is a disadvantage in that not only the same standards are not perfectly available for each person but also labor and time are required.

  To overcome these drawbacks, machine vision systems that automatically detect stains using image processing have become increasingly important. The machine vision system is a system that automatically inspects the surface of the display. It is a system that captures an image of a product to be inspected by a non-human computer and fetches it to a computer, Based on the image processing step and the image processing result, the display surface is automatically inspected through an image analyzing step to determine whether the inspection target product is defective.

Applying a professional - level detection algorithm using such a machine vision system is expected to have a significant effect on time and cost in comparison with human direct spot detection.

In recent years, various algorithms have been proposed for an algorithm for acquiring and processing images using a machine vision system to automate spot blot detection of a display film. Methods for detecting spots include detection in a spatial domain, detection in a frequency domain, and detection in both a spatial domain and a frequency domain.

First, as a method of spatial domain image processing, a method of finding a Region-Mura by a surface fitting method using a polynomial approximation of Wang and Ling has been proposed, but it has a disadvantage in that it can not detect various types of stains. Chen and Chiang proposed a method of detecting stains using threshold values. This method has a limitation in detection of a smear when detecting a smear, if there is a smear and background in the same gray level value. Song et al. Proposed a method of detecting a stain by setting thresholds at various levels after preprocessing through morphological processing, but it has a drawback in that it can not detect various stains. Baek et al. [3] used a method of estimating the background contrast using face matching using polynomial approximation, but there is a disadvantage in that there are many differences in results depending on the threshold value.

In the frequency domain, Chen and Kuo extract the background image at the cutoff frequency after the DCT conversion. The method of finding the stain is effective for background removal, but the stain can be detected by setting the cutoff frequency correctly. C.J.Chen et al. Detected the stain with a CSF (contrast sensitivity function) that mathematically modeled human characteristics that are sensitive to changes occurring at low frequencies rather than high frequencies. However, if the stain is close to the background, the detection power may be lowered. S.J. Kim et al. [5] used wavelet transform to detect blotches, but the speed is fast, but the preprocessing or post-processing process is insufficient to find non-spots. Nakano and Mori proposed speckle detection based on the area improvement method in the wavelet domain and the threshold value in the spatial domain, but it has a disadvantage in that it can not find all kinds of speckles.

  Meanwhile, Xin Bi detects smear by Gabor filter and Level set method. However, it has disadvantage in that smear is detected by using active contour model which searches whole area after Gabor filtering, but active contour method is slow. Two of them proposed an analysis of brightness and dispersion and a method using exponentially weighted moving average (EWMA). However, this method has disadvantages in that it determines the value by trial and error and finds stains.

Accordingly, the present invention proposes a technique for automatically detecting a spot on a display film surface by using morphology image processing and labeling.

Korean Patent Publication No. 10-2014-0006582 (Publication date 2014.01.16.) Korean Patent Publication No. 10-2011-0123512 (published on November 15, 2011).

SUMMARY OF THE INVENTION Accordingly, the present invention has been made to solve the above-mentioned problems occurring in the prior art. Accordingly, an object of the present invention is to provide an image processing apparatus, And detecting a final smear through labeling. The present invention also provides a system and method for automatically detecting a stain defect.

According to another aspect of the present invention, there is provided a system for automatically detecting a spot defect by using morphological image processing and labeling. The system includes an image obtaining unit for obtaining an image of an object, A histogram smoothing unit for obtaining a histogram of an image from which the illumination component has been removed by the preprocessing unit and then expanding the histogram to a whole gray level through histogram smoothing to reconstruct an image; A smoothing candidate generating unit configured to apply a Gabor filter to the image transformed by the processing unit to construct a smoothing candidate region and a smoothing detecting unit configured to detect smoothing using the labeling in the smoothing candidate region formed by the smoothing candidate generating unit .

Meanwhile, the automatic detection method of blotch defects using morphological image processing and labeling of the present invention includes a step of acquiring an image of an object top surface of an object to be imaged by the image acquiring unit, The histogram smoothing unit acquires a histogram of the image from which the illumination component has been removed from the preprocessing unit, and then reconstructs the image by expanding the histogram to the entire gray level through histogram smoothing, Constructing a smear candidate region by applying a Gabor filter to the reconstructed transformed image, and detecting the smear using the labeling in the smear candidate region formed in the smear candidate generating section.

As described above, the system and method for automatically detecting blur defects using morphological image processing and labeling according to the present invention provide the following effects.

1. In the preprocessing process, the background is estimated using the morphological image processing, and the illumination is removed from the original image, and then the algorithm is applied so that the stain can be detected even in the illuminated image.

2. In the step of selecting a candidate area as a stain by labeling, the stain comparison algorithm has an advantage in that the image can be formed only by stain, and thus the detection power is excellent.

3. Detectable results can be obtained without being influenced by various types of stain detection and color such as stain or complex stain.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a block diagram schematically showing a system for detecting an automatic blur defect using morphological image processing and labeling according to a preferred embodiment of the present invention.
2 is a flowchart illustrating a method of automatically detecting a spot defect according to an embodiment of the present invention.
FIG. 3 is a diagram illustrating a process of removing illumination components using a preprocessing morphological image processing technique.
FIG. 4A is a graph illustrating a histogram of a preprocessed image in which the illumination component is removed from the preprocessing unit, and FIG. 4B is a graph illustrating a histogram of the image after histogram smoothing is performed on the preprocessed image.
FIG. 4A is a graph illustrating a histogram of a preprocessed image in which the illumination component is removed from the preprocessing unit, and FIG. 4B is a graph illustrating a histogram of the image after histogram smoothing is performed on the preprocessed image.
FIG. 5 is a diagram illustrating a histogram smoothing process for white and black spots when both white spots and black spots are included on an image. FIG.
6 is a diagram illustrating a Gabor filtering process of a smudge candidate generating unit according to an exemplary embodiment of the present invention.
FIG. 7 is a diagram illustrating a process of labeling using an 8-adjacent relation on an image to which Gabor filtering is applied after morphological preprocessing.
8 is a labeled image of a candidate region of an image after Gabor filtering.
FIG. 9 is a final smear image composed of objects judged to be smear through labeling.
10 is an FPD stain sample image.
Fig. 11 shows the result of detection of the blob by the method according to the embodiment of the present invention using the image of Fig.

In the present specification, as shown in Fig. 1, spot stains are spot stains, spot stains appearing in a region such as a fingerprint are region stains, and foreign stains in the case of stains containing foreign substances such as hair, , And complex stains if there were more than two types of stains.

[Figure 1] Types of stains

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, an auto blotch detection system using morphological image processing and labeling according to the present invention will be described in detail with reference to the accompanying drawings.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a block diagram schematically showing a system for detecting an automatic blur defect using morphological image processing and labeling according to a preferred embodiment of the present invention.

Referring to FIG. 1, an auto-blur defect detection system according to an exemplary embodiment of the present invention includes an image acquisition unit 100 for acquiring an image of an object, An image processing unit 200 for performing necessary processing using image processing software on the image of the object acquired by the acquisition unit 100 and a determination unit for determining whether or not there is a spot defect in the object based on the image processing result of the image processing unit 200 And an image analyzing unit (300).

The image acquiring unit 100 acquires an image of a top surface of an object to be inspected and transmits the acquired image to the image processing unit 200. The image acquiring unit 100 includes an illumination unit arranged to be apart from the upper surface of the object, An image capturing unit 120 that is disposed apart from the upper surface of the object and captures an image using the light reflected from the upper surface of the object, the illumination unit and the image capturing unit are moved on a plane spaced apart from the upper surface of the object, (Not shown) for driving the photoreceptor.

The image processing unit 200 performs a necessary process using the image processing software on the image of the object acquired by the image obtaining unit 100 and transmits the image to the image analyzing unit 300 using morphological image processing The pre-processing unit 210 and the preprocessing unit 220 for removing illumination from the image of the object transmitted from the image acquiring unit 100 distribute the pixel values of the image from which the illumination component is removed to the entire gray level through histogram smoothing And a histogram smoothing unit 220 for reconstructing the image data into a shape that can apply a Gabor filter.

The image analyzing unit 300 determines whether there is a defect by extracting a blur defect from an image of an object transmitted from the image processing unit 200. The image analyzing unit 300 emphasizes a blur component through a garber filter on an image transmitted from the image processing unit 200 A smear candidate generating unit 310 and a smoothing candidate generating unit 310. The smoothing candidate generating unit 310 generates a smoothing candidate region image by assigning a number to each object in the smoothing candidate region image, And a final blob detection unit 320 for reconstructing the image.

Here, a description will be made of a method for automatically detecting blur defects using morphological image processing and labeling of the present invention using the system constructed as described above.

2 is a flowchart illustrating a method of automatically detecting a spot defect according to an embodiment of the present invention.

Referring to FIG. 2, the image acquisition unit 100 acquires a top image of an object to be inspected and transmits the image to the image processing unit 200. For example, the image acquiring unit 100 may photograph the image of the upper surface of the object with the photographing unit 120 and input the photographed image to the image processing unit 200.

Next, the preprocessing unit 210 of the image processing unit 200 receiving the image of the object performs a preprocessing process of removing illumination through morphological image processing when the image of the object includes non-uniform illumination components in one step (Step 1).

The smoothing unit 220 of the image processing unit 200 reconstructs the image by expanding the gray level of the entire range through the histogram smoothing to the image in which the illumination components are removed in two stages (Step 2).

Then, the image analyzing unit 300 determines whether or not there is a defect by extracting a spot defect from the image of the object transmitted from the image processing unit 200 in step 3. The detection of such a speck is performed by the smoothing candidate generating unit 310 by applying a Gabor filter to the image transmitted from the image processing unit 200 to form a smoothing candidate region (Step 3.1) The number of each object in the unevenness candidate region image formed by the unevenly-classified candidate generation unit 310 is numbered to determine whether or not each object is blotted, and the image is reconstructed (Step 3.2).

Hereinafter, each of the above-described steps will be described in detail.

1. Elimination of illumination using morphological image processing

1) Morphological image processing

Morphology is a technique for analyzing and processing shapes in an image. Morphological image processing is used to extract image elements necessary for expressing or describing shapes such as boundaries, blocks, and skeletons of images, and is also used to connect when the boundaries between images are not constant or disconnected. Expansion and erosion are essential operations for morphology, and virtually all morphological image processing is based on these two operations.

Erosion reduces the size of an object in an image constantly using a structural element, thereby reducing the size of the object and enlarging the background. It is therefore useful to remove noise or remove very small objects between objects in the image.

The erosion is expressed by Equation (1) when there are sets A and B in the space of Z ² . Here, A is an object and B is a structure element. This equation implies that the erosion of A by B is a set of all points Z contained in A by B moved by Z. Figure 2 shows the erosion process. Figure 2a is the original image, Figure 2b is the structural element, Figure 2c is the erosion operation to the original image, and Figure 2d is the final erosion image.

[Equation 1]

[Figure 2] Erosion Operation

Dilation can result in enlarging the object size and reducing the background by removing the protrusion inside the object and increasing the protrusion outside. It is used to fill the hole-like space inside the object or connect short-cut areas.

Expression when expansion A is in the space of Z ² and B is in the space of Z ² is expressed by Equation (2). Figure 3 shows the expansion process. Figure 3a is the original image, Figure 3b is the structural element, Figure 3c is the process of performing the expansion operation with the structural element in the original image, and Figure 3d is the final expansion image.

&Quot; (2) "

[Figure 3] Expansion operation

Opening is an expansion operation after an erosion operation, removing the convexity and breaking the narrow connection to smooth out the outline of the image. The shape and size of the object is preserved and is called an elimination operation because it removes the protrusion and the narrow connection.

The heat of assembly A by the structural element B is the expansion of the result by B after the erosion operation of A by B as shown in Equation 3. [ Figure 4 shows the opening process. Fig. 4a shows the original image, Fig. 4b shows the structure element, Fig. 4c shows the result of erosion operation on the original image, and Fig. 4d shows the image after the expansion process.

&Quot; (3) "

[Figure 4] Open computation

Closing refers to performing an erosion operation after an expansion operation as opposed to an open operation. Like the opening operation, filling the concave part or the small hole smooths the outline part of the image, and the shape and size of the object are preserved. It is also a filling operation because it also serves to fill small holes and gaps.

Similarly, the closing of set A by structure element B is to erode the result by B after the expansion operation of A by B. The close operation can be expressed by Equation (4). Figure 5 shows the close process, Figure 5a shows the original image, Figure 5b shows the structure element, Figure 5c shows the result of the expansion operation on the original image, Figure 5d shows the result of the close operation after the erosion process, to be.

&Quot; (4) "

[Figure 5] Close operation

2) Pretreatment method using morphological image processing

In the smear detection method according to an embodiment of the present invention, when an illumination component is included, an illumination component is determined as smear and an erroneous result may be obtained. Therefore, a preprocessing process is performed to remove the illumination component before applying the smear detection algorithm . Hereinafter, the preprocessing method of the smear image through the open and close operations will be described.

If two images having the same illumination condition, that is, the image including the smear and the image without the smear, are divided, the pixel value having the same gray scale component is treated as the same gray scale component, so that the effect of removing the illumination component can be obtained. , It is necessary to use a method of estimating and removing illumination from the image itself in order to remove the illumination component.

In the embodiment of the present invention, a method of removing an illumination component of an image using a morphological image processing technique is used. The preprocessing unit 210 generates a candidate image identical to the original image, estimates an illumination component, and subtracts the original image from the estimated image to remove illumination. More specifically, when the image is applied to an image using an open operation and a close operation using a structural element among the morphological image processing methods, the illumination can be estimated from the original image. The effect of elimination can be achieved and the original stain component remains intact.

FIG. 3 shows a process of removing illumination components using a preprocessing morphological image processing technique. As shown in FIG. 3, when an open operation is performed on the original image transmitted from the image obtaining unit 100, a small protrusion or a noise piece is removed from the original image, and a close operation is performed on the image as shown in FIG. 3B The small holes and gaps are removed. FIG. 3C is a diagram showing an estimation of illumination components after the open-close operation. FIG. 3D is an image representing a reconstructed image obtained by performing a subtraction between the original image and the image of FIG. 3C.

2. Histogram Smoothing

The histogram smoothing unit 220 extends histogram smoothing to a gray level of the entire range in order to efficiently apply the Gabor filter of the smoothing candidate generating unit 310, which will be described later, to the image from which the illumination component has been removed by the preprocessing unit 210 Reconstruct the image.

FIG. 4A is a graph illustrating a histogram of a preprocessed image in which the illumination component is removed from the preprocessing unit, and FIG. 4B is a graph illustrating a histogram of the image after histogram smoothing is performed on the preprocessed image.

As shown in FIG. 4, when the histogram smoothing unit 220 performs a difference operation with the image obtained by estimating the illumination component in the original image, the background components have the same value, The value obtained by subtracting the grayscale value of the background component is obtained as shown in FIG. 4B. Therefore, the histogram smoothing is performed to expand the entire gray level to emphasize the stain as shown in FIG. 4B.

5 illustrates histogram smoothing processes for white and black spots when both white spots and black spots are included on the image. Referring to FIG. 5, the histogram smoothing unit 220 calculates a histogram If there are two white spots and one black speckle, if only one side is preprocessed, only one of the black spots or white speckles will be found. Histogram smoothing can be done in two ways. Thus, the first image is subjected to the histogram smoothing process as it is, and the second image is smoothed after the image is inverted.

3. Stain defect extraction

3.1 Gover  Construct a candidate region for unevenness using a filter

1) Gabor filtering

The Gabor filter is a filter that is similar in frequency and orientation to the human visual system and is suitable for representation or identification of the pattern. The 2D gauge filter is defined by the 2D Gaussian function and the complex function as shown in equation (5). In Equation (5), u ₀ and v ₀ are specific spatial frequencies, and g (x, y) is a Gaussian function.

&Quot; (5) "

는 표준편차이다.The Gaussian function is expressed by Equation (6). here,

Is the standard deviation.

&Quot; (6) "

Since the 2D Gabor filter function is a complex function, it can be divided into a real part and an imaginary part as follows. The real part of the Gabor filter is shown in Equation 7 and is shown in Fig. The imaginary part is expressed by Equation 8 and is shown in FIG.

&Quot; (7) "

[Figure 6] Spatial Domain Gabor Filter

&Quot; (8) "

[Figure 7] Frequency domain Gabor filter

2) Configuration of unevenness candidate region using Gabor filter

The blur candidate generation unit 310 of the image analysis unit 300 according to the embodiment of the present invention uses real gauge filtering to emphasize the blurring, And constitute a candidate region. The formula of the real number Garber filter is as shown in equation (9).

&Quot; (9) "

는 주파수,

는 회전각, u₀는

, v₀는

이다.In Equation 9,

Is the frequency,

Is the rotation angle, u ₀ is

, v ₀

to be.

On the other hand, the smudge candidate generating unit 320 performs Gabor filtering on the two images generated after the preprocessing process, and reconstructs the two images into one image through an AND operation.

6 is a diagram illustrating a Gabor filtering process of a smudge candidate generating unit according to an exemplary embodiment of the present invention. Referring to FIG. 6, the smear candidate generating unit 320 first displays the images divided into two parts by histogram smoothing after morphological preprocessing, and expresses the two images as an image through an AND operation.

3.2 Candidate Area Labeling  After stain selection

1) Labeling

The labeling method is to divide each object in the binarized image by 4-connected or 8-connected by using the connection relation. to be.

When labeling, 4-connectivity and 8-connectivity are shown in Fig.

A pixel p in (x, y) on the image has two horizontal lines whose coordinates are (x + 1, y), (x-1, y), (x, y + The neighbors have two vertical neighbors, and the four neighboring sets of p are denoted by N ₄ (p), as shown in Figure 8a. The four diagonal neighbors of pixel p are (x + 1, y + 1), (x + 1, y - 1), It is called 8-neighbors of p and ₈ (p) together with the horizontal component.

If two pixels p on the image and neighboring pixels N ₄ (p) are 4-adjacent to each other and p and neighboring pixels are adjacent to N ₈ (p) 8-adjacent). As shown in Fig. 8c, the 4-adjacency relation can be said to be 8-adjacency with 4-adjacency, but Fig. 8d can be said to be 8-adjacency without 4-adjacency.

[Figure 8] 4-Connectivity and 8-Connectivity

The labeling results are different according to the 4-adjacent relation and the 8-adjacent relation. If the labeling is performed using the 4-adjacent relation, the connection is maintained only for the vertical and horizontal components. When labeling is performed using the 8-adjacent relationship, pixels in the vertical, horizontal, and diagonal directions are included in the adjacent relationship, and thus are represented in two steps as shown in FIG. 9b.

[Figure 9] Labeling with 4-connectivity and 8-connectivity

2) Label candidate regions

In the speckle detection unit 320 according to the embodiment of the present invention, the candidate region binarized after Gabor filtering is finally determined to be speckle using labeling and determined.

FIG. 7 is a diagram illustrating a process of labeling using an 8-adjacent relationship on an image to which Gabor filtering is applied after morphological preprocessing, and FIGS. 7A to 7D respectively show images of the following steps 701 to 704.

As shown in FIG. 7C, if the average value of the first labeling area and the average value of the surrounding area are 97.5 to 100% identical, the background is recognized as the background and excluded from the labeling area. If the average value is less than 97.5% . The final labeling image is shown in FIG. 7D, and the formula (10) represents a criterion for determining the labeling object.

≪ Step 701 > Gabor filtering image candidates are labeled using 8-connectivity.

≪ Step 702 > When the number of pixels of the entire labeling object is 3 pixels or less, it is not determined as a spot, but is deleted from the labeling image and excluded from the smear candidate.

≪ Step 703 > For each labeled object, a surrounding value comparison formula is applied as shown in Equation (10) to determine whether the background is a spot or background. The region determined as background is excluded from the labeled image. Construct the final image.

≪ Step 704 > The final image is composed of objects determined to be stains.

&Quot; (10) "

FIG. 8 is a labeled image of a candidate region of an image after Gabor filtering, and FIG. 9 is a final smear image composed of objects determined to be smear through labeling. If it is determined that the number of the background blur is a background component and the number of remaining blur is determined to be blur, then the final blur image is composed only of the number of blur, as shown in FIG.

Table 1 shows the types of image stains used in the experiment and the detection results according to the embodiment of the present invention.

For detection and evaluation of automatic detection method of stain defect using morphological image processing and labeling, each image was classified by size and stain type. All images were composed of 410,401 pixels, totaling 168,100 pixels. As shown in Table 1, the samples used to calculate the detection rate were a total of 200 stain samples, such as spot stain, region stain, foreign stain, and complex stain. A spot is a speck in the form of a speckle, and a region speckle is a speckle with a fingerprint. The foreign substance is an image containing a hair-like foreign substance, and complex stains are used in combination of two or more kinds of stains. The images were classified into two types of white and black on the image.

Stain type Number of experimental images Number of detected images Detection rate spot 32 26 81.25% region 135 134 99.3% 유산 14 14 100% complex 19 17 88.2% sum 200 191 95.5%

As shown in Table 1, in the method of detecting a stain according to an embodiment of the present invention, nine images were not detected in a total of 200 images, and the detection rate was as high as 95.5%.

In this specification, the objective reliability was evaluated by applying the Semu index (SEMI D31-1102, 2002.) to the SEMU for luminance Mura in FPD image quality inspection. The Semu Index was created in 2002 to provide an objective assessment of the evaluation of FPD stains in SEMI. The Semu index is expressed as the ratio of C _JND , the minimum identification difference that can be perceived by human vision, and C _x , the average contrast of the stain.

Here, C _JND is inversely proportional to the area Sx of the detected smear and is expressed by Equation (11).

&Quot; (11) "

The Semu is defined by the ratio of C _x and C _JND to Equation (12). The difference C _x between the detected stain and the background is proportional to the area S _x of the detected stain, and as the difference between the detected stain and background lightness and the detected stain area increases, the semu value increases, As the difference in contrast between the background and the area of the detected stain decreases, the Semu value also decreases.

&Quot; (12) "

Where C _JND is the minimum contrast for spot detection, C _x is the average of the measured stain, and S _x is the surface area of the measured stain.

As shown in Table 2 and Table 3, the method according to the present invention and the conventional method (LC Chen, CC Kuo, Automatic TFT-LCD mura defect inspection using discrete cosine transform-based background filtering and just noticeable The results of the Semu evaluation are compared with the results of the difference quantification strategies, Meas. Sci. Technol., vol.19, pp. 110, 2008). In the same experimental conditions as the conventional method, the difference between the detectable stain and the minimum contrast of the background was evaluated to evaluate the Semu value.

The experimental image size of the conventional method was 256256, the size of the blot was about 1,800 pixels, and 10 blotches of difference in contrast between the average value of the blot and the background were used. On the other hand, in order to estimate the background value of the experimental image of the conventional method, it is confirmed that the gray level is between 120 and 125 by substituting into the expression (12).

In the experimental example of the present invention, in order to ensure consistency with the conventional method, the size of the stain on the size 410410 was used as an experimental image with an average gray level of about 1,800 pixels. 10A and 10B illustrate an FPD smear sample image, in which FIG. 10A shows a smear image with a pixel value of 135 with a contrast difference of 10 between a smear and a background, and FIGS. 10B to 10J show a smear image with a difference of light and dark .

Fig. 11 shows the result of detection of the blob by the method according to the embodiment of the present invention using the image of Fig. As shown in FIG. 11G, the speckle is detected from the difference of intensity 4, but the speckle is not detected in FIG. 11H. Therefore, the detection method according to the embodiment of the present invention detects speckles from the difference of intensity 4.

Table 2 shows the reliability according to an embodiment of the present invention, and Table 3 shows the reliability of the conventional method (LC Chen, CC Kuo, Automatic TFT-LCD mura defect inspection using discrete cosine transform-based background filtering and just noticeable difference quantification strategies, Meas Sci. Technol., Vol. 19, pp. 110, 2008).

Table 2 shows the Semu evaluation according to an embodiment of the present invention. The larger the background image is, the more difficult it is for the human eye to recognize it, so the value of | C _x | becomes smaller and the value of | C _x | becomes larger for a darker image. C _JND represents the minimum contrast difference that can be perceived by the human eye as a ratio to the size of the stain. Therefore, the larger the size of the smear, the smaller the value of the C _JND item. Therefore, it is easy to detect a stain with a larger staining value such as | C _x | / C _JND . Table 3 shows the results of the semu evaluation of the difference between the unevenness of the conventional method and the background darkness difference, and the method according to the embodiment of the present invention shows an improved result as compared with the conventional method.

On the other hand, at the gray level 120 to 125 of the background image in Table 3, | C _x | The item was expected to be close to 4 at about 4.16, but the Semu evaluation result of the conventional method is | C _x | The item is 3.07, and the semu value is about 3 and 4. It can be seen that the detection of the smear is not performed properly at the difference of intensity 5.

Contrast difference

(%)

(%)

(%) Semu evaluation 10 8.1366 0.8958 9.0830 success 9 7.3274 0.9080 8.0701 success 8 6.5375 0.9056 7.2191 success 7 5.7384 0.9034 6.3521 success 6 4.9416 0.8990 5.4969 success 5 4.1453 0.8978 4.6172 success 4 3.3465 0.8972 3.7298 success 3 2.8447 1.6414 1.7330 failure 2 2.8540 1.6414 1.7387 failure One 2.8636 1.6414 1.7446 failure

Contrast difference

(%)

(%)

(%) Semu evaluation 10 8.32 0.88 9.3 success 9 7.42 0.88 8.39 success 8 6.61 0.88 7.48 success 7 5.79 0.88 6.55 success 6 4.78 0.88 5.39 success 5 3.07 0.89 3.44 success 4 0.52 1.22 0.43 failure 3 0.48 1.28 0.31 failure 2 0.37 1.28 0.29 failure One 0.18 1.5 0.12 failure

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it is to be understood that the invention is not limited to the disclosed exemplary embodiments, but, on the contrary, is intended to cover various modifications and equivalent arrangements included within the spirit and scope of the invention.

100: image acquisition unit 200: image processing unit
210: preprocessing unit 220: histogram smoothing unit
300: image analyzing unit 310: smear candidate generating unit
320:

Claims (10)

An image acquiring unit for acquiring an image of an object,
A preprocessor for removing an illumination component from the image acquired by the image acquiring unit using morphological image processing; and a processor for acquiring a histogram of the image from which the illumination component has been removed in the preprocessing unit and then expanding the histogram to the entire gray level through histogram smoothing An image processing unit having a histogram smoothing unit for reconstructing an image;
A smoothing candidate generating unit for applying a Gabor filter to the image converted by the image processing unit to form a smoothing candidate region; and a smoothing detecting unit for detecting smoothing using the labeling in the smoothing candidate region formed by the smoothing candidate generating unit An automatic detection system for blotch defects using morphological image processing and labeling including image analysis unit. The method according to claim 1,
The pre-
An illumination operation is applied to an image by using an open operation and a close operation using a structural element, and morphological image processing for removing illumination components by subtracting the illumination component from the image, system. The method according to claim 1 or 2,
The histogram smoothing unit may include:
If the image of the object includes white and black speckles, the white speck image is distinguished from the black speckle image, and the black speckle is automatically detected by inverting the histogram and then performing smoothing. system. The method according to claim 1,
Wherein the smear candidate generating unit comprises:
An automatic detection system for blotch defects using morphological image processing and labeling using the real number Gabor filtering of Equation 1 below.
(Equation 1)

However,

Is the frequency,

Is the rotation angle, u ₀ is

, v ₀

to be. The method according to claim 1,
The above-
If the number of pixels of the entire labeling object is 3 pixels or less, it is not determined as a spot and is excluded from the unevenness candidates. If the number of pixels of the entire labeling object is 3 pixels or less, The automatic detection system of stain defect using morphological image processing and labeling.
(Equation 2)

The image acquiring unit acquiring an image of the object upper surface,
Removing an illumination component from the image acquired by the image acquisition unit using a preprocessing additive morphological image processing,
The histogram smoothing unit acquires a histogram of the image from which the illumination component is removed from the preprocessing unit, and then reconstructs the image by expanding the histogram to a whole gray level through histogram smoothing,
Forming a smudge candidate region by applying a Gabor filter to the transformed image reconstructed by the histogram smoothing unit;
And detecting a blob by using a labeling in a blob candidate region formed in the blob candidate generating unit by the blob detecting unit. The method of claim 6,
Wherein the step of removing the illumination component comprises:
Estimating an illumination component by applying an open operation and a close operation using a structure element to an image, and
And subtracting the illumination component from the image to subtract the illumination component from the image, thereby automatically detecting the blotch defect using the morphological image processing and labeling. The method according to claim 6 or 7,
Wherein the step of reconstructing an image by expanding the entire gray level through the histogram smoothing comprises:
If the image of the object includes white and black speckles, the white speck image is distinguished from the black speckle image, and the black speckle is automatically detected by inverting the histogram and then performing smoothing. Way. The method of claim 6,
Wherein the step of constructing the smear candidate region comprises:
A method of automatic detection of blur defects using morphological image processing and labeling using the real number Gabor filtering of Equation (3) below.
(Equation 3)

However,

Is the frequency,

Is the rotation angle, u ₀ is

, v ₀

to be. The method of claim 6,
Wherein the step of detecting the smear comprises:
Labeling the unevenness candidate region using 8-connectivity,
A step of excluded from the smear candidates without judging that the number of pixels of the entire labeling object is 3 pixels or less,
Determining whether each of the labeling objects is a speckle or background through Equation 4, and excluding the background from the speckle candidates; and automatically detecting the speckle defect using the morphological image processing and labeling.
(Equation 4)

Images