Stereo image comfort quality evaluation method and system based on convolutional autoencoder

Abstract

The invention provides a method and a system for evaluating the quality of stereoscopic image comfort based on a convolutional self-encoder, which comprises the following steps: a convolutional self-encoder model is established and unsupervised learning training is carried out; the basic features of left and right stereoscopic images are extracted by using the trained convolutional self-encoder model; by using the extracted basic features, the depth features corresponding to the stereoscopic images of left and right viewpoints are extracted; the dimension of the extracted basic features and depth features is reduced, and the dimension- reduced basic features and depth features are spliced and expanded into one-dimensional vectors; the obtained one-dimensional vector is processed, and the stereo image comfort model is obtained by learning. The invention introduces a convolution self-encoder to carry out initial feature extraction, effectively removes redundant information of an image, is very efficient in feature extraction, and matches the depth extraction algorithm and the SVR regressor, so that the effect is improved remarkably.

Description

technical field [0001] The invention relates to the field of image comfort quality evaluation, in particular to a method and system for stereo image comfort quality evaluation based on a convolutional autoencoder. Background technique [0002] In recent years, stereoscopic (3D) video industries such as stereoscopic movies and stereoscopic televisions have developed rapidly, and 3D video resources have increasingly entered people's daily lives. While improving the viewing experience, it will also cause the audience to experience discomfort such as dizziness, nausea, and chest tightness. In order to improve the discomfort of watching stereoscopic images and further enhance the viewing experience, a large number of researches on the comfort quality evaluation methods of stereoscopic images have emerged as the times require. Stereoscopic image comfort quality evaluation methods can be divided into subjective evaluation and objective evaluation. The former is very time-consuming...

AI Technical Summary

Problems solved by technology

While improving the viewing experience, it will also cause discomfort such as dizziness, nausea, and chest tightness to the audience.

Images