Research And Application Of TFT-LCD Defect Detection Algorithm Based On Image Background Reconstruction And Level Set Method

Thin film transistor-liquid crystal display(TFT-LCD) nowadays is the most common display,in order to improve the quality of TFT-LCD products, more accurately and efficiently detect the defective products, foreign manufacturers have developed the automated inspection equipment to replace manual inspection. But in domestic, many manufacturers still rely on manual inspection; although some domestic companies have tried to develop automatic inspection equipment, because of the detection algorithms have missed and false detection problems, the developed equipments are difficult to be applied in practice.To overcome these shortcomings of current detection algorithms, this paper proposes a TFT-LCD defect detection algorithm based on image background reconstruction and level set method, and builds an experimental platform to verify the proposed algorithm, finally, the algorithm is applied to automatic inspection equipment for the practical test. The specific research contents are as follows:(1) Hardware system design of inspection platform, the hardware architecture is designed and the selection program is proposed according to the requirements of display detection.(2) For the problem of false detection caused by non-uniform background image, this paper proposes a background image reconstruction algorithm based on the prejudgment of defect region, to eliminate the influence of pixels in defect area on background image reconstruction, and obtain the background image with high accuracy. After the background image is obtained, using the original image subtract the background image to reduce the inhomogeneity of image background.(3) For the problem that the existing active contour model based display defects segmentation method is easy to cause missed detection because of the insensitivity to small defects. In this paper, an active contour model based on gradient and region information is proposed and use level set method to evolve contour curve to segment defects, the algorithm can deal with the inhomogeneous background situation and is very sensitive to small defects, it effectively improves the detection accuracy.(4) To accelerate the algorithm the GPU parallel processing technology is used. this paper also designs software structure and function of the inspection platform, and realizes the application of the algorithm on the inspection platform.(5) Through the experiment to make a comparesion between the proposed algorithms and other scholars proposed TFT-LCD defects detection algorithms, the algorithm was evaluated from the robustness, precision and detection rate and false positive rate of Mura defects, and the experimental results were quantized and analyzed. The results show that the proposed algorithm has high detection accuracy, low false detection rate, and the detection effect is better than the compared algorithms. Finally, the proposed algorithm is transplanted to the automatic inspection device and tested by the actual application, the detection rate of point and line defects was 100%, the detection rate of Mura defect was 93.56%, the false positive rate was 5.16%, the test results show that the algorithm has reached the practical application requirements.