PROJECT TITLE :

Multi-Channel and Multi-Model-Based Auto encoding Prior for grayscale image restoration

ABSTRACT:

Low-level image processing's long-standing challenge is image restoration (IR). To get aesthetically appealing outcomes, learning excellent image priors is critical. As an image prior for solving the IR problem, we develop a multi-channel and multi-model denoising autoencoder network in this research. An initial prior is initially constructed using the network's training data from RGB images, and then it is applied to single-channel, grayscale IR tasks. Auxiliary variables are utilised in the iterative restoration process to incorporate higher-dimensional network-driven prior information. The idea of weighted aggregation also proposes a multi-model technique to improve network stability and prevent it from becoming stuck in local optima. this. Experimental evidence shows that the suggested approach is effective, robust, and capable of restoring grayscale photographs to their original quality.