No-reference stereopair quality assessment based on singular value decomposition

Abstract

Abstract Singular value decomposition (SVD) is an effective mathematical tool that has attracted considerable interest in the area of image quality assessment (IQA). Although it has been widely used for full-reference image quality prediction, its capacity to measure visual distortions for no-reference (NR) IQA has not been explored in depth. Here we propose a new SVD-based NR 3D stereopair quality assessment model, named SSQA, to amend this limitation. In the proposed method, the influences of various distortions to energy and structure of single views are considered by seeking changes of singular values and singular vectors. In particular, we quantify the correlation between left and right views with the difference of singular values and mutual information of them. A set of “quality-aware” features are extracted from the left and right views. We use a machine learning method to predict the quality of images. We test our algorithm on four 3D image databases. The experimental results show that the performance of SSQA model is competitive with existing efficient methods on both symmetric and asymmetric distortions.