The present invention relates to a GPU-based TFT-LCD Mura defect detection method comprising the following steps: (1) a binary quadratic regression diagnostic model based on studentized residuals, based on original image data; Obtaining the bi-linear equation regression background image data; (2) obtaining an influence amount of each group of data points with respect to a conformance value based on the original image data and the data of the pair quadratic regression background image; (3) obtaining a new pixel group based on the unique points and the influence points in the original image data excluding the influence amount; (4) setting a pair N-order polynomial surface approximation model based on the new pixel group and obtaining the pair N-order background image data; (5) acquire a residual image R based on the pair N-order background image data and the original image data, and perform threshold segmentation using a threshold value on the residual image to obtain a threshold-segmented image (6) morphology processing is performed on the threshold-divided image to obtain an eroded and dilated image, and realization of effective partitioning for a non-uniform Mura defect is performed.
展开▼