Machine Vision Based Liquid Crystal Display Mura Defect Detection

Mura defect,a kind of display defect widely exists in LCD,has the characteristics of blurry contour,low contrast,and unfixed position and shape.Currently,artificial detection is primarily performed in Mura defect detection.However,it works inefficiently and its application is affected by external conditions and subjective factors,and thus it fails to meet the requirements set for product quality and production efficiency in LCD production.Therefore,it is an urgent need to develop a kind of machine vision detection method which is fast,accurate and free from interference in the detection process.Owing to the characteristics of Mura defects and the complicated backgrounds of the defect images,both the traditional threshold segmentation method and edge detection method fail to detect Mura defects effectively.Focusing on solving the difficulties existed in Mura defect detection,this paper makes a research on Mura defect machine vision detection based on three key points,namely,image texture background suppression,defect segmentation and defect quantization.The research work is as follows:The method of image texture background suppression is investigated.For the problem that the detection of Mura defects is interfered by repetitive texture background of LCD images,on the basis of the comparisons of different textured background suppression method,this paper conducts a research on Gabor wavelet filtering theory and the design of filters methods,designs the 16 channel real value Gabor wavelet filter bank,and adopts the fusion method which is targeted at the filter multi-channel output sub-images,achieving effective suppression on image texture background.Mura defect segmentation method is studied.Aiming at solving the problem that the traditional image segmentation algorithm fails to correctly segment Mura defects,an improved C-V(Chan-Vese)model is proposed.The improved model simplifies the image data driven force of the traditional C-V model,speeds up the segmentation process,adds an energy item which is related to the gray average both inside and outside of the contour curve,diminishes the influence on detection result caused by the asymmetric luminance of the image background,and together with the level set method,segmenting Mura defects accurately at the segmenting speed that is six times as fast as the traditional C-V model.Quantitative approach for the Mura defects is researched.On the condition that defects are segmented correctly,Mura defects are quantized through the extraction of the defect area and average contrast,and the location and shape of Mura defects are ascertained through the method proposed in this paper,this method first converts the original defect image to the image which has only two gray values,and then works out the value of the center of mass,the area,and the enclosing rectangle of the defect area,achieving the target of ascertaining the location and shape of Mura defects.Finally,Mura defect machine vision detection system is constructed,and a test experiment is performed on 50 LCD samples with Mura defects,achieving 94 percent accuracy with the Mean Time to Detection(MTTD)controlled in 20 seconds.The result shows that the proposed machine vision based LCD Mura defect automatic measurement method is available,achieving the design target.