Deep Image Compression Perceptual Reconstruction Network Based on Channel Attention
Bin Xie, Huaqin Cai · 2023
This paper proposes a depth image compressed sensing image reconstruction based on channel attention to solve the problem that the image reconstruction algorithm based on deep learning cannot well recover the original signal from the measured value when the sparsity distribution of the signal is uneven. construct a network. The proximal gradient descent algorithm (PGD) and the deep unrolling network (DUN) are combined to build a compressed sensing image reconstruction model. A channel attention mechanism is added to the encoder and decoder of the proximal mapping module (PMM) to help the model focus. Focus on important information, improve the accuracy of signal recovery, and better restore the original signal. At the same time, Charbonnier loss and edge loss have been improved to better adapt to different tasks and sampling rate requirements. Experiments show that this model algorithm can effectively improve the reconstruction accuracy of image compressed sensing.