Multi-scale input dual decoder Blind Motion Deblurring

Abstract

Image deblurring is a classic problem in low-level computer vision, which aims to recover a sharp image from the input blurred image. At present, there are some problems in the mainstream network, such as complex model architecture, difficult training and poor effect of restoring sharp image details. To solve the above problems, an improved model based on conditional generation adversarial network is proposed. The model makes the generator network structure lightweight, the encoder stage fuses multi-scale feature information, and the dual decoder is used to learn the complementary residual information in different directions respectively, so as to further improve the texture details of the deblurs image. Finally, performance tests are performed on the GoPro dataset. After experimental evaluation, it is proved that compared with other similar methods, the proposed algorithm has better restoration ability, higher detail restoration degree of blurred image, higher peak signal-to-noise ratio (29.45dB) and higher structural similarity (0.93). It can complete the motion blur task of image blind better.