A no-reference evaluation method for depth image quality based on natural scene statistics

Abstract

The present invention proposes a no-reference evaluation method for depth image quality based on natural scene statistics, including steps: (1) collecting a group of depth images, and dividing the collected depth images into two parts, one part is a training image, and the other part is Test image; (2) Extract feature parameters on different scales for each depth image. The extraction steps are: extract the edge area of ​​the scale image through edge detection, and find the scale image gradient magnitude and Gauss-Ra in the edge area. The distribution of the Placian operator, and respectively use Weber distribution and asymmetric Gaussian distribution to establish the distribution function models of the two, and use the parameters of the two models as the characteristic parameters of the depth image; (3) finally, use the characteristic parameters of the training image Perform random forest model training to generate an objective quality score evaluation model; input the characteristic parameters of the test image into the objective quality score evaluation model to obtain the objective quality score of the test image.

Description

technical field [0001] The invention relates to the field of image quality evaluation, in particular to a method for evaluating depth image quality without reference based on natural scene statistics. Background technique [0002] The existing quality evaluation methods are divided into subjective evaluation methods and objective evaluation methods. Although the subjective evaluation method is the most accurate evaluation result, it is time-consuming and laborious, and it is not feasible in actual operation. Therefore, it is of great significance to design an objective quality evaluation method. The quality evaluation methods that can be used to evaluate the depth image mainly include traditional image quality evaluation methods and early quality evaluation methods for depth images. These methods are introduced and analyzed one by one below. [0003] 1. Traditional image quality evaluation methods: There are many traditional algorithms. The full parameter algorithm evalu...

AI Technical Summary

Problems solved by technology

[0005] Judging from the existing algorithms, the traditional image quality evaluation methods, whether full-parameter or non-parametric algorithms, have not been designed for the particularity of depth map distortion, resulting in poor evaluation accuracy

However, the current algorithms for evaluating the quality of depth maps need to use undistorted texture maps for evaluation, which is not easy to implement.

Images