Non-reference evaluation method for quality of virtual viewing angle synthetic video

Abstract

The invention discloses a non-reference evaluation method for quality of a virtual viewing angle synthetic video. The method comprises two modules: a module 1: carrying out twinkled distorted region extraction based on double detections on a to-be-evaluated viewing angle synthetic video so as to extract out all twinkled blocks in the video; and a module 2: carrying out singular value decomposition on the extracted twinkled blocks in each frame and blocks in a previous frame, which are matched with the twinkled blocks, then carrying out normalization on all singular values by utilizing a maximum singular value of each twinkled block, obtaining a distance between each twinkled block and an average value of the normalized singular values of the block matched with the twinkled block, and using the distance as twinkling distorting intensity of the block. An F norm of distorting intensity of each twinkled block is used as a quality score of the block, a sum of the quality scores of all the twinkled blocks in each frame is used as a quality score of a current frame, and an average value of the quality scores of all the frames is used as a final quality score of the video. According to the non-reference evaluation method disclosed by the invention, quality of the virtual viewing angle synthetic video can be evaluated accurately.

Description

technical field [0001] The invention relates to an objective visual quality evaluation method for virtual viewing angle synthesis, in particular to a reference-free evaluation method for virtual viewing angle synthesis video quality. Background technique [0002] Currently, multi-view and free-view videos have attracted widespread attention. The realization of these two types of video requires a large number of perspective images, and due to the limitation of network bandwidth and cost, it is impossible to complete all perspectives through cameras, so it is necessary to use virtual perspective synthesis technology to synthesize new perspectives from multiple known perspectives. Perspective image. The most commonly used in the existing virtual perspective synthesis technology is the depth map-based rendering method, that is, the Depth Image Based Rendering (DIBR) synthesis technology. The existing DIBR synthesis technology will cause holes in the foreground or background ed...

AI Technical Summary

Problems solved by technology

Although this method realizes the quality evaluation of virtual view synthesis video, the performance still needs to be improved

Images