Uneven Illumination Correction in Whole Slide Imaging using Pix2Pix

This study proposed an effective real time method for uneven illumination correction in whole slide imaging based on deep learning and pix2pix structure. We developed a customized network and provided an in-depth analysis of the effectiveness of important hyperparameters such as the bottleneck size. We also investigated the performance of common pixel-wised, feature matching, and perceptual loss functions (MAE, MSE SSIM, and LPIPS) in reconstructing images’ details. The result of the proposed network has been compared to the most highlighted traditional methods and recently proposed deep structure. The uneven illumination has been eliminated effectively in the results.