Automatic detection of Mura defect in TFT-LCD based on regression diagnostics

This paper proposes a computationally efficient Mura defect detection method that is constructed based on regression diagnostics using the prediction error sum of squares (PRESS) residuals and an image estimation procedure for automatic Mura inspection of thin film transistor liquid crystal display (TFT-LCD) devices. The gray-level data of the input image is estimated by a linear model and then the PRESS residuals are calculated to filter Mura regions out. After image dilation, the threshold value is determined for detecting the non-uniform brightness or darkness areas in TFT-LCD by means of examining every pixel in the image. The experimental results of several test images returned by using the proposed method and an existing method in the literature are used to evaluate the performance of effectiveness and efficiency for Mura detection. It has been found that the method proposed in this paper is very swift in processing time and also returns competitive Mura detection performance in comparison to the exiting method.