CN107633507A - LCD defect inspection methods based on contour detecting and characteristic matching - Google Patents

Abstract

The invention discloses an LCD defect detection method based on contour detection and feature matching. The steps are as follows: first collect m standard LCD display screen images and average them to establish a standard library, re-acquire and update the library every 2 minutes; then collect the images to be tested LCD display image; then register the standard image and the image to be tested, using the method based on contour detection and feature matching; then perform weighted average fusion on the registered image to be tested and the standard image to obtain a new image to be tested ; Afterwards, local adaptive threshold segmentation is performed on the fused image to be tested and the standard image; finally, the defect is detected by the subtraction method, and the type and location of the defect are counted by the least circumscribed rectangle method. The invention can detect LCD defects in real time and with high precision.

Description

LCD Defect Detection Method Based on Contour Detection and Feature Matching

technical field

The invention relates to the field of LCD display defect detection, in particular to an LCD defect detection method based on contour detection and feature matching.

Background technique

Liquid crystal display devices are widely used in various household appliances and instruments, and have made great progress. Due to the complicated production process and the influence of the surrounding environment, the display screen is prone to defects. Therefore, the detection of LCD display defects is of great significance for improving the production process of LCD display screens and improving its product quality. Commonly used methods include artificial visual inspection, electrical parameter inspection, and automatic optical inspection. The former two are generally used to detect macroscopic defects, but cannot detect microscopic defects. In traditional artificial visual inspection, the resolution of the human eye is not high, and errors in detection may occur. Inspection is highly subjective, and long-term work will cause visual fatigue, resulting in low stability. It is difficult to guarantee the accuracy of quality inspection, and it cannot become a unified inspection standard. Automatic optical inspection has developed rapidly due to its advantages of non-contact and high performance. Many scholars have done research on the automatic detection of LCD display defects, but among a large number of methods, some do not overcome the influence of light, some are sensitive to image rotation, some require the background of the object under test to be simple, and some cannot detect defect information etc., and there is basically no way to detect defects that are close to the target background.

The core of LCD defect detection is image registration. The higher the registration accuracy, the higher the detection accuracy. There are mainly two methods based on region and feature. Currently, the feature-based method is the most common. The traditional SIFT algorithm is robust to image rotation and scale change, but it has a large amount of computation and high time complexity. The SURF algorithm based on SIFT sacrifices accuracy for speed. Later, many scholars improved on the basis of SIFT and SURF. However, the accuracy of image registration of most methods is average, and the scope of application is relatively limited; the anti-interference ability to image brightness changes is not high; the rotation angle that can be processed is small; all must The mismatch removal algorithm is used, and the time complexity is high; all methods use affine transformation in the registration process, and affine transformation is a special case of transmission transformation, which can only handle two-dimensional space rotation and translation. If image micro-distortion occurs 3D state, affine transformation will be wrong.

Contents of the invention

The purpose of the present invention is to provide a simple, fast, and high-accuracy LCD defect detection method to meet the needs of various LCD detection markets.

The technical solution that realizes the object of the present invention is: a kind of LCD defect detection method based on profile detection and feature matching, comprises the following steps:

Step 1. Collect and average m standard LCD screen images to obtain a standard template image and establish a standard library, where m is a positive integer;

Step 2, collecting the image of the LCD display screen to be tested;

Step 3. Register the standard template image in step 1 with the image to be tested in step 2, and obtain the image to be tested after registration by using a method based on contour detection and feature matching;

Step 4, performing fusion processing on the standard template image in step 1 and the image to be tested obtained in step 3 to obtain the image to be tested after fusion;

Step 5, performing threshold segmentation on the standard template image in step 1 and the image to be tested obtained in step 4, respectively, to obtain two threshold images;

Step 6: Process the two threshold maps obtained in step 5 by using the subtraction method to detect defects.

Further, N≥5 in step 1.

Further, the process of collecting and averaging N standard images described in step 1 needs to be repeated every 2 minutes to update the standard library.

Further, the method based on contour detection and feature matching in step 3, the specific process is as follows:

(1) Fill and expand the image to be tested and the standard template image in step 2 to obtain two w×h rectangular images, and w≠h, where w is width and h is height;

(2) Perform global threshold segmentation on the image to be tested after (1) extension to obtain a binary image;

(3) Perform contour detection on the binary image in (2), remove contours less than 0.5×l or greater than l, leave the contour of the image target area, and save it as a preliminary contour map, where l is the rectangle in (1) perimeter;

(4) Carry out contour detection from the top layer to the bottom to the preliminary contour image in (3), save it as a 2D point set, and set up the minimum circumscribed rectangle of the top layer point set;

(5) analyze and calculate the minimum circumscribed rectangle obtained in (4), and determine the absolute value of the angle of rotation θ and the counterclockwise four vertex coordinates between it and the horizontal axis in the counterclockwise direction;

(6) Determine whether the rotation angle of the image to be tested in step 2 relative to the standard template image is positive/negative based on the coordinates of the four vertices obtained in (5), and then select -θ or 90 ^° -θ for the image to be tested in step 2 Perform affine transformation to obtain a preliminary registration map;

(7) Matching the preliminary registration image in (6) and the standard template image based on feature points to obtain the image to be tested after registration.

Further, the fusion process in step 4 specifically adopts the method of weighted average fusion, and the formula used is:

B'(M,N)=c ₁ A(M,N)+c ₂ B(M,N)

In the formula, A is the standard image, B is the image to be tested after registration, and B’ is the image to be tested after fusion, and the size is M×N,

Weighting factor: In the present invention, c ₁ =0.38 and c ₂ =0.62 are selected.

Further, the threshold segmentation in step 5 specifically adopts local adaptive threshold segmentation, and its sliding window size is 9×9.

Compared with the existing method, the present invention has significant advantages in that: (1) the present invention needs 2 minutes to automatically update the standard library, which avoids the impact of illumination changes on the detection results to a certain extent, and improves the detection accuracy; (2) in the present invention The registration method based on contour detection and feature matching is faster and more accurate than traditional feature matching; (3) the weighted average fusion and local adaptive threshold segmentation in the present invention can detect defects similar to the background, The missed detection rate is reduced.

Description of drawings

FIG. 1 is a flow chart of the LCD defect detection method based on contour detection and feature matching in the present invention.

Fig. 2 is a flow chart of the registration process based on contour detection and feature matching in the present invention.

Figure 3 is an embodiment diagram of the LCD defect detection method based on contour detection and feature matching in the present invention, wherein Figure (a) is a standard template figure, Figure (b) is a figure to be detected, and Figure (c) is a fusion to be tested Figures, Figures (d) and (e) are the threshold segmentation diagrams of Figures (a) and (c), respectively, Figure (f) is the result of defect detection, and Figure (g) is the statistical result of defect information.

detailed description

The present invention will be described in further detail below in conjunction with the accompanying drawings and specific embodiments.

A kind of LCD defect detection method based on contour detection and feature matching of the present invention comprises the following steps:

Step 1. Collect and average N standard LCD screen images to obtain a standard template image and build a standard library, wherein N is a positive integer; said N≥5. The process of collecting and averaging N standard images needs to be repeated every 2 minutes to update the standard library.

Step 2, collecting the image of the LCD display screen to be tested;

Step 3. Register the standard template image in step 1 and the image to be tested in step 2, and use the method based on contour detection and feature matching to obtain the image to be tested after registration; the specific process is as follows:

Step 3-1. Fill and expand the image to be tested and the standard template image in step 2 to obtain two w×h rectangular images, and w≠h, where w is width and h is height;

Step 3-2, performing global threshold segmentation on the image to be tested after step 3-1 extension, to obtain a binary image;

Step 3-3. Perform contour detection on the binary image in step 3-2, remove contours smaller than 0.5×l or larger than l, leave the contour of the image target area, and save it as a preliminary contour map, where l is step 3 The perimeter of the rectangle in -1;

Step 3-4, performing contour detection from the top layer to the bottom on the preliminary contour image in step 3-3, saving it as a 2D point set, and establishing the minimum circumscribed rectangle of the top layer point set;

Step 3-5, analyze and calculate the minimum circumscribed rectangle obtained in step 3-4, and determine the absolute value of the rotation angle θ and the counterclockwise four vertex coordinates between it and the horizontal axis in the counterclockwise direction;

Step 3-6. According to the four vertex coordinates obtained in step 3-5, determine whether the rotation angle of the image to be tested in step 2 relative to the standard template image is positive/negative, and then select -θ or 90 ^° -θ for step 2. Perform affine transformation on the image to be tested to obtain a preliminary registration image;

Step 3-7: Perform feature point-based matching on the preliminary registration image in step 3-6 and the standard template image to obtain a registered image to be tested.

Step 4. Carry out fusion processing on the standard template image in step 1 and the image to be tested obtained in step 3 to obtain the image to be tested after fusion; the fusion process specifically adopts weighted average fusion, and the formula used is:

B'(M,N)=c ₁ A(M,N)+c ₂ B(M,N)

In the formula, A is the standard template image, B is the image to be tested after registration, and the size is M×N, where M and N are both positive integers, B' is the image to be tested after fusion, and the weighting coefficient is:

The weighting coefficients are preferably: c ₁ =0.38, c ₂ =0.62.

Step 5. Perform threshold segmentation on the standard template image in step 1 and the image to be tested obtained in step 4 to obtain two threshold images; threshold segmentation specifically uses local adaptive threshold segmentation, and its sliding window size is 9×9 .

Step 6: Process the two threshold maps obtained in step 5 by using the subtraction method to detect defects.

The weighted average fusion and local self-adaptive threshold segmentation in the present invention can detect defects similar to the background and reduce the missed detection rate.

A more detailed description follows.

In conjunction with Fig. 1, the LCD defect detection method based on contour detection and feature matching of the present invention comprises the following steps:

Step 1. Collect m standard LCD display images and average them to obtain a standard template image and build a standard library. This process needs to be repeated every 2 minutes to update the standard library to overcome the influence of light changes;

Step 2, collecting the image of the LCD display screen to be tested;

Step 3. Register the standard template image in step 1 with the image to be tested in step 2, and adopt a method based on contour detection and feature matching. The specific process is combined with FIG. 2 to obtain the image to be tested after registration;

Step 4. Perform weighted average fusion processing on the standard template image in step 1 and the image to be tested obtained in step 3 to obtain the image to be tested after fusion. The formula used is:

B'(M,N)=c ₁ A(M,N)+c ₂ B(M,N)

In the formula, A is the standard image, B is the image to be tested after registration, and the size is M×N, and B' is the image to be tested after fusion, and the weighting coefficient is: The present invention selects c ₁ =0.38, c ₂ =0.62;

Step 5. Carry out local adaptive threshold segmentation on the standard image in step 1 and the fused image to be tested obtained in step 4 respectively. The sliding window size is 9×9 to obtain two threshold images;

Step 6. The two threshold maps obtained in step 5 are processed by subtraction method to detect defects, and the type and position of defects are counted by the least circumscribed rectangle method.

Compared with the traditional method, the detection method of the present invention is not only fast but also has a high accuracy rate, the accuracy rate can reach 98.667%, and has a good application prospect.

The specific description will be given below in conjunction with the embodiments.

Example

Combined with Figure 3, the method is:

(1) After collecting 10 standard LCD images and averaging them, the standard template image is obtained as shown in (a), and stored in the standard library;

(2) Collect the image to be tested as shown in (b);

(3) Using the method based on contour detection and feature matching, the pictures (a) and (b) are registered, and the picture (c) is obtained after weighted average fusion, and the fusion can largely overcome the influence of illumination;

(4) Carry out local adaptive threshold segmentation on pictures (a) and (c) to obtain pictures (d) and (e). This threshold segmentation method can basically distinguish any defects and reduce the missed detection rate;

(6) Perform the subtraction method to detect the defects in Figure (d) and Figure (e) as shown in Figure (f), and use the minimum circumscribed rectangle method to count the type and location of defects as shown in Figure (g);

The method of the invention is simple, fast, and has a high accuracy rate, the accuracy rate can reach 98.667%, meets the needs of various LCD defect detection markets, and has a good application prospect.

Claims (7)

1. an LCD defect detection method based on profile detection and feature matching, is characterized in that, comprises the following steps: Step 1. Collect and average N standard LCD screen images to obtain a standard template image and establish a standard library, where N is a positive integer; Step 2, collecting the image of the LCD display screen to be tested; Step 3. Register the standard template image in step 1 with the image to be tested in step 2, and obtain the image to be tested after registration by using a method based on contour detection and feature matching; Step 4, performing fusion processing on the standard template image in step 1 and the image to be tested obtained in step 3 to obtain the image to be tested after fusion; Step 5, performing threshold segmentation on the standard template image in step 1 and the image to be tested obtained in step 4, respectively, to obtain two threshold images; Step 6: Process the two threshold maps obtained in step 5 by using the subtraction method to detect defects. 2. The LCD defect detection method based on contour detection and feature matching according to claim 1, characterized in that N≥5 in step 1. 3. The LCD defect detection method based on contour detection and feature matching according to claim 1, characterized in that, the collection of N standard images described in step 1 is averaged, and this process needs to be repeated every 2min to update the standard library . 4. the LCD defect detection method based on contour detection and feature matching according to claim 1, is characterized in that, in step 3, based on the method for contour detection and feature matching, concrete process is as follows: Step 3-1. Fill and expand the image to be tested and the standard template image in step 2 to obtain two w×h rectangular images, and w≠h, where w is width and h is height; Step 3-2, performing global threshold segmentation on the image to be tested after step 3-1 extension, to obtain a binary image; Step 3-3. Perform contour detection on the binary image in step 3-2, remove contours smaller than 0.5×l or larger than l, leave the contour of the image target area, and save it as a preliminary contour map, where l is step 3 The perimeter of the rectangle in -1; Step 3-4, performing contour detection from the top layer to the bottom on the preliminary contour image in step 3-3, saving it as a 2D point set, and establishing the minimum circumscribed rectangle of the top layer point set; Step 3-5, analyze and calculate the minimum circumscribed rectangle obtained in step 3-4, and determine the absolute value of the rotation angle θ and the counterclockwise four vertex coordinates between it and the horizontal axis in the counterclockwise direction; Step 3-6. According to the four vertex coordinates obtained in step 3-5, determine whether the rotation angle of the image to be tested in step 2 relative to the standard template image is positive/negative, and then select -θ or 90°-θ for the rotation angle of step 2 Perform affine transformation on the image to be tested to obtain a preliminary registration image; Step 3-7: Perform feature point-based matching on the preliminary registration image in step 3-6 and the standard template image to obtain a registered image to be tested. 5. the LCD defect detection method based on contour detection and feature matching according to claim 1, is characterized in that, fusion process in step 4, what specifically adopt is weighted average fusion, and used formula is: B'(M,N)=c ₁ A(M,N)+c ₂ B(M,N) In the formula, A is the standard template image, B is the image to be tested after registration, and the size is M×N, where M and N are both positive integers, B' is the image to be tested after fusion, and the weighting coefficient is: c ₂ =1-c ₁ . 6. The LCD defect detection method based on contour detection and feature matching according to claim 1, wherein the threshold segmentation in step 5 specifically adopts local adaptive threshold segmentation with a sliding window size of 9×9. 7 . The LCD defect detection method based on contour detection and feature matching according to claim 5 , wherein the weighting coefficients c ₁ =0.38 and c ₂ =0.62.