Quality evaluation method without reference image for viewing angle synthesis

Abstract

The invention provides a quality evaluation method without a reference image for viewing angle synthesis. According to the method, through analysis of characteristics of two types of distortion, features are designed to quantify distortion in an synthetic image, wherein first, distortion and texture non-naturality brought by edge damage of the image are quantified, and corresponding features are extracted; and then a machine learning method is used to integrate the features, and therefore a quality evaluation model capable of evaluating distortion brought in the whole synthesis process is trained. The method overcomes two defects in existing methods, wherein the first defect is that through the existing methods, only one type of distortion in the synthesis process can be evaluated, but through the method, the two types of distortion in the whole synthesis process can be effectively evaluated; and the second defect is that most of the existing methods are full-reference methods, that is, only when an original undistorted image is provided can a distorted image be subjected to quality evaluation through the existing methods, but the method is a reference-free method and has wider application prospects.

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 image quality evaluation method for viewing angle synthesis. Background technique [0002] View synthesis is to use texture image and depth image to synthesize a new view image. View synthesis technology is widely used in fields such as multi-view video and free-view TV [1]. The quality evaluation of the synthesized views can quantitatively evaluate the quality of the view synthesis technique, and can also be used to optimize the synthesis technique. Furthermore, the quality evaluation of view-synthesized images directly affects the success of these applications. Therefore, the quality evaluation for view synthesis is of great significance. [0003] Distortions in synthetic perspective fall into two main categories. The first category is the traditional distortion brought by the acquisition, processing and tr...

AI Technical Summary

Problems solved by technology

[0004] The above existing synthetic image quality evaluation methods have the following defects: First, each method is designed for one type of distortion, while ignoring the other distortion brought in the perspective image synthesis process

In reality, the original image is often not available, which restricts the application of existing full-reference methods

Images