Convolutional neural network (CNN)-based stereoscopic-image comfort degree evaluation method

Abstract

The invention belongs to the field of image processing, and aims to provide technology that a convolutional neural network (CNN)-based stereoscopic-image comfort degree evaluation method is adopted toobtain an evaluation score of an image, a subjective evaluation value of a human on the stereoscopic image is better fitted, and a stereoscopic-image comfort degree can be very well predicted. Therefor, the adopted technical solution in the invention is the convolutional neural network-based stereoscopic-image comfort degree evaluation method. Information of a left view, a right view and a disparity map of the stereoscopic image is synthesized into one image to use the same as input of a network, human perception is simulated through the convolutional neural network to process the obtained image, weight coefficients are obtained through training, and finally, human eye vision significance features are utilized to carry out weighting on image block comfort values to form a comfort value ofthe overall image. The method is mainly applied to image processing.

Description

technical field [0001] The invention belongs to the field of image processing and relates to the improvement and optimization of an objective evaluation method for the comfort degree of a stereoscopic image. Specifically, it involves a three-dimensional image comfort evaluation method based on convolutional neural network. Background technique [0002] Stereoscopic images are subject to external interference during the process of acquisition, compression, storage, transmission, and display, which leads to image degradation and seriously affects people's viewing experience. Therefore, evaluating the comfort level of stereo images is one of the main problems to be solved urgently in the field of stereo imaging technology. [0003] Full-reference stereoscopic image quality evaluation plays a very important role in the whole objective stereoscopic image quality evaluation. Many excellent full-reference stereo image quality assessment algorithms have been proposed, and they can...

AI Technical Summary

Problems solved by technology

[0005] The convolutional neural network uses the ReLU activation function to solve the problem o

Images