Stereo image objective quality evaluation method integrating manifold and binocular features

Abstract

The invention provides a stereo image objective quality evaluation method integrating manifold and binocular features. Based on human eye manifold perception, an orthogonal local preserving projection algorithm is used to acquire a matrix after dimensionality reduction and albefaction from a natural scene plane image for training to acquire an optimal mapping matrix. In order to improve the accuracy and stability of evaluation, image blocks which are not important to visual perception are removed. The manifold feature vector of selected image blocks is extracted through the optimal mapping matrix. Structure distortion of a distortion image is measured through manifold feature similarity. The influence of the image brightness variation on a human eye is considered. The brightness distortion of the distortion image is calculated based on the mean of the image blocks. After the manifold similarity and the brightness similarity are acquired, a binocular contention model is used to carry out linear weighting on the quality values of left and right viewpoint images to acquire the quality value of a distortion stereo image. An evaluation result is highly consistent with the subjective perception evaluation result of the human eye.

Description

technical field [0001] The invention relates to a method for evaluating the quality of a stereoscopic image, in particular to an objective method for evaluating the quality of a stereoscopic image combining manifold features and binocular characteristics. Background technique [0002] Quantitative evaluation of stereoscopic image quality is a challenging problem in the field of image processing. Stereoscopic images differ from planar images in that stereoscopic images contain two different viewpoints. When a person watches a stereoscopic image, the human visual system (HVS) does not process the left-viewpoint image and the right-viewpoint image separately, but forms a pair of eyes after a complex binocular fusion and competition process. This fused cyclopean map not only depends on the differences in individual stimuli, but also depends on the geometric relationship between the two viewpoints. Therefore, the quality of a stereoscopic image is not only related to the qualit...

AI Technical Summary

Problems solved by technology

However, the second method needs to consider two issues. First, since the real depth / disparity map is not always available, the second method generally evaluates the quality of stereoscopic image depth perception when estimating the depth map, so , the accuracy of the depth / disparity estimation algorithm may greatly affect the evaluation performance; secondly, the depth / disparity information may not correlate with the 3D perception quality, which has been confirmed in the article of Kaptein et al., who used in different subjective experiments in different Blurred images of the same object at depth, found that depth does not affect perceived image quality to some extent in 3D displays

[0005] Despite the aforementioned problems, the above two methods still achieve good quality assessment performance in stereoscopic image quality assessment for symmetric distortion.

However, if the left and right viewpoints contain different degrees or types of distortion (also known as asymmetric distortion stimuli), then the above two methods perform poorly in quality assessment

Asymmetric distortion makes the stereoscopic image quality evaluation problem more challenging, mainly because the quality of the cyclopean image synthesized by the human eye is related to the distortion type and distribution of the left and right viewpoints

Images