Fast Image Restoration Using A Simplified Matrix-Type Recurrent Neural Network
Liqing Huang, Youshen Xia · 2022 15th International Congress on Image and Signal Processing, BioMedical Engineering and Informatics (CISP-BMEI) · 2022
Recently, a matrix-type recurrent neural network with two-layers structure has been successfully applied for the matrix-variable optimization model of image restoration, since it can effectively strengthen the computation speed of vector-variable optimization methods. This paper employs a simplified matrix-type neural network with one-layer structure to further improve the computational performance. First, a matrix-variable regularization optimization model is considered for image detail perseveration. Then a simplified matrix-type neural network is introduced for fast dealing with matrix-variable regularization optimization model. Finally, illustrative examples show that the simplified matrix-type recurrent neural network-based algorithm is more effective than the existing matrix-type neural network with two-layers structures in terms of computational performance, and has faster speed than the vector-type neural networks.