Reaserch Of Mura Defects Detection Method Based On B-spline Surface Fitting And Snake Model

As a kind of outward appearance defects that widespread exist in Thin Film Transistor Liquid Crystal Display, which is known as TFT-LCD, Mura defects are the most difficult defects to be detected accurately. Because of its character of low contrast, no sharp edges and uneven brightness, Mura defects are so hard to be recognized by Machine Vision that domestic and international leading TFT-LCD manufactures still have to rely on Artificial Visual Inspection(AVI) to detect such defects. However, the result of AVI cannot meet the Flat Panel Display industry’s requirements of highly automated, standardized, objective, unified manufacture principle. Moreover, AVI is low efficient. Therefore, a swift and accurate Mura defects detection method using Machine Vision is urgent needed.Based on the research status quo of Mura defects Machine Vision detection, a set of rapid and accurate Mura defects Machine Vision detection method that consist of three key parts, image preprocessing, background suppression and Mura defects segmentation, is proposed. In the part of image preprocessing, irrelevant background that exist outside the TFT-LCD region would be captured as well during the process of TFT-LCD image acquisition, meanwhile, TFT-LCD regional issues such as image tilt and geometrical deformation would occur inevitably. To avoid such issues’ interference to the accuracy of detection result and extract the TFT-LCD area integrallty, Hough transform is employed to obtain the tilt angle then rotation formula is deployed to adjust the tilt region using acquired tilt angle, thereafter, interesting TFT-LCD regional image corner is extracted based on Harris corner detector. Therefore, TFT-LCD region is cropped from the original image in accordance with the extracted corner of TFT-LCD matrix region. Thereafter, interesting TFT-LCD region is filtered by the low-pass Butterworth filter. In the process of background suppression, to solve the problem of accuracy of Mura detection caused by uneven brightness and complex background, an adjusted background suppression method which adds a fairing item is proposed based on the traditional least square bi-cubic B-spline surface fitting that always weaken Mura defects when suppressing background because of its too high fitting accuracy. In order to improve the performance of proposed method, bi-cubic B-spline function is resolved by one-dimensional formulas under the principle of product function, divided fitting of the original TFT-LCD image and data compression of the control points which can further increase the efficiency of the algorithm are deployed at the same time. According to the characters that made it difficult to be detected correctly, a Snake mode based on the mean curvature motion is proposed to segment Mura defects. In the process of Mura segmentation, to figure out the problem of the diverse shape and a variable number of Mura defects region, adaptive initial curve is employed and the level set function is introduced. Similarly, in order to increase the performance of the proposed method, the termination condition is set in accordance with the length of the zero level set of binary curves to decrease the number of iterations and AOS format solving method is used to solve the Snake model. Eventually, a Mura defect detection system using Machine Vision is established based on the proposed Mura detection methods. In order to quantize the segmented Mura defects and grade the tested TFT-LCD, the SEMU standard which had been already internationally accepted by the Industry is employed.Finally, to verify the accuracy and high performance of the proposed detection methods and Machine Visual system, experimental detection is conducted on the samples that selected randomly from TFT-LCD production line and 99 out 100 TFT-LCD samples are successfully detected, furthermore, the average detection time is no more than 30 s,which mean the efficiency and accuracy are both higher than the current AVI and Machine Vision detection methods. The experimental results show that the proposed Mura defection automatic detection method and established Machine system can meet the requirement of accurate, stable and efficient detection of Mura defects.