An Adaptive Thresholding Based Method to Locate and Segment Defects on LCD Panels

Real time defect detection plays a vital role in the liquid crystal display (LCD) manufacturing industry. The different shapes and types of defects poses a great challenge to accurately segment the defects. To achieve the objective this paper presented a LCD panel defect segmentation method. The proposed method constituted three parts: (1) preprocessing of the defective images was necessary to normalize and minimize the noise presented in the images. (2) Then, an image was represented in gray levels whose frequencies were computed using histogram function. But inconsistency was present in the gray levels due to strong illumination gradients. Therefore, the probability of intensity value at all locations of the image was computed to obtain the total variance. The obtained total variance as a thresholding factor in the image was used to separate the image into background and foreground. A thresholding factor was computed adaptively by averaging all the pixel values for the segmentation purpose. (3) Lastly, a binarized defect image was obtained from matching the ground truth image to the obtained threshold image by using pixel by pixel multiplications. Experimental results and comparative study showed that the presented approach was able to correctly locate and segment various shapes of defects with better metrics values.