HEVC optimization method combining decompression effect and super-resolution

Abstract

The invention discloses an HEVC optimization method combining a decompression effect and super-resolution. The method mainly comprises the following steps: at an encoding end, an original video is firstly subjected to down-sampling and then is compressed, and therefore the data volume of the video can be reduced; at a decoding end, a double-network structure method is adopted, a mapping relationship between a downsampling video and a compressed low-resolution video under different quantization parameters is trained firstly, so that a compression artifact can be weakened, then a mapping relationship between a quality-enhanced compressed video and an original video is established, and finally, a quality-improved video frame can be obtained at an output end. Experimental results show that ina low bit rate segment, the rate distortion performance of the coded video can be effectively improved, and a better visual effect can be obtained.

Description

technical field [0001] The invention relates to video compression and super-resolution reconstruction technology, in particular to an HEVC coding optimization method combining decompression effect and super-resolution, belonging to the field of image communication. Background technique [0002] In image and video transmission, in order to adapt to the limitation of bandwidth and facilitate people's storage, it is usually down-sampled and compressed. Image and video compression standards generally use block-based hybrid coding methods. The main purpose of these compression technologies is to compress a large amount of redundant information to adapt to the bandwidth limitation of communication channels. The idea of ​​these technologies is to quantize the transform domain coefficients to reduce the amount of data, but it will cause block effects, thereby reducing the quality of images and videos, and the combination of downsampling and compression techniques often makes compres...

AI Technical Summary

Problems solved by technology

However, most of the existing methods based on deep learning are aimed at a single downsampling step, and there is still room for improvement in the quality of compressed low-resolution videos.

Images