Multi-description compressed image enhancement method based on residual recursion compensation and feature fusion

Abstract

The invention discloses a multi-description compressed image enhancement method based on residual recursion compensation and feature fusion, belongs to the field of image quality enhancement, and solves the problem of different degrees of compression distortion of an image compressed by a multi-description coding method, especially the problem of serious structure splitting artifacts of a side decoded image. The method comprises the following steps: firstly, designing a residual recursive compensation network as a low-resolution feature extraction network of a side path and a middle path, and more effectively extracting two description decoding image features with the same content and different details by using a parameter sharing strategy; secondly, enabling the multi-description side feature up-sampling reconstruction network to adopt a network part layer parameter sharing strategy, so that the size of a network model is greatly reduced, and the generalization ability of the network is improved. Meanwhile, a multi-description middle-path feature up-sampling reconstruction network is used for performing deep feature fusion on two side-path low-resolution features and a middle-path low-resolution feature, so that efficient multi-description compressed image quality enhancement is realized, and the performance of the method is superior to that of a plurality of deep learning image enhancement methods such as ARCNN, FastARCNN and DnCNN.

Description

Technical field [0001] The present invention belongs to the field of image quality enhancement, particularly relates to a residual based on recursive compensation feature fusion and image enhancement multiple description compression method. Background technique [0002] While the existing communication system can provide a wide bandwidth, but in crowded places such as live concerts, football venue, student housing complex there will be network congestion phenomenon. In addition, the extreme poor in remote areas typically have limited resources, communications equipment, which will result in packet loss has a great probability of occurrence. Although the conventional image compression standard can achieve efficient compression, but does not guarantee reliable data transmission. Unlike single-described compression, multiple description coding (MDC) is divided into a plurality of source description, and are transmitted through a different channel, then the image received at the rece...

AI Technical Summary

Problems solved by technology

The network can well solve the problems of large storage space and high computational complexity caused by the large size of the existing deep learning model, and can solve the problems of varying degrees of distortion in compressed images, especially the severe structural splitting artifacts in side-channel decoded images. shadow phenomenon

Images