A Multi-Feature Fusion Convolution Neural Network for Image Compression Artifacts Reduction
Bingqi Lin, Feifeng Wang, Chen Jing, Linlin Chen · 2021 IEEE International Conference on Networking, Sensing and Control (ICNSC) · 2021
For effective image storage and transmission, lossy compression is normally used during the coding process, which introduces artifacts and destroys the quality of images inevitably. Though some convolutional neural network (CNN) based image compression artifacts reduction methods have been proposed, the reconstructed quality is not ideal. In this paper, an end-to-end multi-branch convolution neural network is designed for removing artifacts brought by JPEG compression. Three feature extraction branches are proposed to extract main texture features, multi-scale features and enhancement features of compressed image. With the proposed network, multi-scale features of compressed image are fully concerned, artifacts of adjacent block boundaries can be reduced effectively. Experimental results show the effectiveness of the proposed network. In particular, the PSNR-B index of the proposed network is higher than state-of-the-art methods, which indicates the better artifact reduction performance of the proposed network.