No-reference Stereoscopic Image Quality Assessment Based on Visual Saliency Region

Abstract

Stereoscopic image quality assessment (SIQA) plays an important role in the development of 3D image system. In view of the problem for existing SIQA methods cannot effectively extracted features of stereoscopic image and the dimension for the related features data extracted from stereoscopic image is too large. In this paper, we proposed a no-reference stereoscopic image quality assessment method combined with visual saliency regions and wavelet transform. The method firstly combines the distorted image pairs into two separated cyclopean images by using Gabor filtering and the SSIM-based stereoscopic algorithm. Then, detect the visual significant region for the distorted image pair, the two synthetic cyclopean image and the corresponding depth map respectively and segment those images into patches. Make a wavelet decomposition and obtain the phase amplitude and gradient features of the wavelet subband as stereoscopic image features. Finally, we use SVR to establish the mapping relationship between the features of stereo image quality and DMOS. The experimental results show that compared with the existing state-of-the-art no-reference stereoscopic image quality assessment (NR SIQA) methods, the proposed model can maintain good consistency with human subjective perception.