P‐40: A Novel LCD De‐Mura Algorithm Based on Deep Learning

Since its proposal in 1968, LCD has held an importa nt position in the display field due to its mature tech nology and low cost advantages[1]. In order to elimin ate mura defects in VA‐LCD panels, the traditional D emura method requires taking multiple images of diff erent graylevel, fitting a full graylevels brightness cur ve, and then calculating compensation values. The lo ng photography time affects the production line capac ity. Therefore, the paper proposes a Demura method based on deep learning, which only requires taking a single graylevel image as the input, using the comp ensation values of traditional schemes as learning lab els,and using U‐shaped neural network for model trai ning, to predict the compensation values of different binding point graylevels. While greatly improving effic iency, it can achieve the effect of industry standard s chemes. The experimental results show that after ado pting this scheme, the number of photos taken can b e reduced from 7 to 1, the Demura efficiency can be improved by 51%, the compensation data prediction error is 0.3%, and the panel uniformity is consistent with the traditional schemes;