Study On The Methods Of Automatic Optical Inspection For The Mura Defect On TFT-LCD Displays

The Manufacturing process of TFT-LCD displays has many working procedures，and its structure is complicated, therefore all TFT-LCD inevitably contains a variety ofvisual defects, these defects may influence human health if they are clear enough.Nonuniform illumination on the displays is one of the most common visual defects, thisphenomena was called Mura. At present, Mura detection mainly rely on human eyes,this approach is subjectivity and inefficiency, increasingly unable to meets therequirements of high-volume and large area for TFT-LCD displays production. It isimperative to research on automatic detection method. The difficulty of automaticdetection is amorphous, low contrast, blurry contour and complicated background, etc.This article makes studies of automatic optical inspection algorithm aimed at Mura,this method is independent of human eyes. Considering of the characteristic of pendingimages and the difficulty of Mura detection, the entire algorithm is design into threeparts: image pre-processing, defect segmentation and feature extraction, the core of themethod is background suppression. For image noise, Butterworth low-pass filter is usedto denoise. For objects which not belong to the TFT-LCD displays, global thresholdmethod and Hough transform based on geometric features of regions is proposed. Foruneven brightness, a brightness adjustment method based on standard template image isproposed, the Mura defects are reversed at the same time. The background of TFT-LCDis treated as a three-dimensional surface, hence bicubic B-spline fitting method is usedto reconstruct and suppress the background. After processing by a series of algorithms,the nonuniform illumination region may be destroyed, by an appropriate combination ofa variety of operations of Mathematical Morphology, it is possible to repair the largerconnected region, and inhibit the smaller connected region at the same time. Thepurpose of feature extraction is to determine whether the Mura defects is existed, andrate it. The selected characteristics include contrast, area, perimeter, location and otherparameters according to actual need and human visual mentality.This algorithm follows the SEMI standard, using SEMU to determine whether aTFT-LCD display containing Mura defects. Experiment on45TFT-LCD displays, and 43were successfully detected, the success rate is95.56%. The experiment reveals thatthe algorithm proposed in this article reached the predetermined design requirements inaspects of effectiveness, real-time and stability, etc.