3D-DWT cross-band statistics and features for No-Reference Video Quality Assessment (NR-VQA)

Abstract

This paper presents a robust, novel, and computationally efficient noise estimation-based NR-VQA model. It uses four novel sub-band features; namely cross-band statistics, sub-band kurtosis, sub-band energy ratios, and natural video statistics in three-dimensional discrete wavelet transform (3D-DWT) domain. As the HVS shows high sensitivity towards the structural and textural information, a three-dimensional gray level co-occurrence matrix (3D-GLCM) is used to select top 25% video blocks having maximal textural information. We validate the performance of the proposed method on five publicly available and widely used video quality databases. Extensive experiments show that the proposed model yields considerably better and robust performance than other contemporary state-of-the-art NR-VQA methods at lower computational costs.