Detection of Water-Stains Defects in TFT-LCD Based on Machine Vision

Abstract

TFT-LCD is an important part of the mobile phone, and the defect detection of it requires a lot of manpower and material resources. Common TFT -LCD defects include point defect, line defect and Mura defect. The water-stains defects are a common Mura defect. This paper presents an efficient algorithm for detecting water defects. The first step is to strengthen the detect features by band-pass filter. Then, Sobel edge detection operator is used to enhance the edge of the abnormal defect. The backlight of the mobile phone screen may not be uniformly distributed, and the enhanced image takes on an irregular texture background, so the screen can be divided into several small blocks. After that, a background assessment is made on each small piece to roughly locate where the defect area is. Finally, to make a second judge on the small pieces of interest, a SVM classifier is trained. Through many tests in the mobile screen production workshop, the accuracy of the algorithm is 95.9%, meeting the requirements of industrial inspection.