Image Super-resolution with Enhanced Channel Attention Residual Network

Abstract

Convolutional neural networks (CNN) have been successful in the field of image super-resolution, achieving state-of-the-art results. While Transformer-based neural networks are becoming increasingly popular, CNN-based deep neural networks still have great potential in this field. In this study, we optimized and adapted Residual Channel Attention Networks (RCAN) for image super-resolution. Specifically, we proposed an enhanced channel attention residual block and introduced a multi-spectral channel attention mechanism into the network, which leads to an improved performance of the attention mechanism. Experimental results demonstrate that our proposed Enhanced Channel Attention Residual Network (ECARN) achieves significant performance improvement compared to previous models. Our approach represents a promising direction for further research into image super-resolution using CNN-based deep neural networks with attention mechanisms. (Abstract)