Compressed video stream re-encoding method based on deep learning and saliency perception

Abstract

The invention provides a compressed video stream recoding method based on deep learning and significance perception. The compressed video stream recoding method comprises the following steps: constructing and training a compressed domain video image significance detection deep learning model; inputting a compressed video image X to be recoded into the compressed domain video image saliency detection deep learning model CDVNet trained in the step 1; decoding the X part of the compressed video image to be recoded by using the compressed domain video image saliency detection deep learning model CDVNet; performing video image recoding by using an HEVC coding technology and combining the updated quantization parameter of each coding unit. According to the method, saliency feature extraction based on a compressed domain is adopted; data information obtained by partial decoding is used for saliency detection in a compressed code stream, the defect that in the prior art, feature extraction andsaliency detection can be carried out only after all compressed videos are decompressed to a pixel domain in saliency detection based on the pixel domain is overcome, and the method has the advantages of being small in calculation amount and low in time consumption.

Description

technical field [0001] The present invention relates to the technical field of video image processing, in particular to a method for re-encoding compressed video streams based on deep learning and saliency perception in the technical field of video image compression. Background technique [0002] The systematization and standardization of video image compression technologies, such as JPEG, JPEG2000, H.264 / AVC, HEVC, etc., have made it normal to store and transmit massive video image data in compressed form. Limited by objective conditions such as business, confidentiality, or bandwidth, in some compressed image data applications, it is necessary to provide or transmit image data with different resolutions. For example, the transmission of high-definition video images on a network with limited bandwidth needs to reduce the resolution and transmission rate; The grades of hyperspectral images of soldiers are also different. In addition, the display accuracy of various display...

AI Technical Summary

Problems solved by technology

This method has achieved a certain effect on the compression ratio, but this "full decompression and full compression" structure cannot make good use of the information obtained from the first compression, which not only wastes computing and cache resources, but also takes a long time to compress, making it difficult to do real-time processing

Images