Maximum a posteriori based unsupervised learning for images deblurring with cauchy noise

In image restoration with a known blur kernel, Gaussian noise is commonly assumed, and corresponding removal methods have been extensively studied. In contrast, less concentration has been devoted to Cauchy noise, which exhibits strong impulsive characteristics. In this work, we propose an unsupervised image restoration framework based on a convolutional neural network with dropout to simultaneously remove Cauchy noise and blur. We derive from a maximum a posteriori (MAP) estimation formulation to constraint the proposed network, the final restored image is obtained by averaging multiple predictions generated by the network. We evaluate the proposed method on three benchmark datasets using PSNR and SSIM as evaluation metrics. Compared with seven competing methods, experimental results demonstrate the proposed approach achieves superior performance in both quantitative metrics and visual quality.