Abstract
Independent component analysis (ICA) is used to detect the mura regions with varying sizes and brightness levels before thresholding, then individually analyzed the mura regions in order to avoid unnecessary background effect. Defects detection is performed by partitioning test image into overlapping sub-windows and Classifying each sub-window as normal or mura region by comparing the difference of spatial distance between ICs of defective and non-defective. During the experimental process, a median filter and a high-pass filter are also respectively used to filter out the noise and enhance mura gray intensity. In this research, we developed ICA to achieve off-line learning and on-line detection.
