A New No-reference Color Image Quality Assessment Metric in Wavelet and Gradient Domains
Besma Sadou, Atidel Lahoulou, Toufik Bouden · 2018
Most of image quality assessment methods are designed for grayscale images on a single domain. Therefore, they do not sufficiently make use of image color and multi-domain information that could provide valuable insights on the human visual mechanisms. In this paper, we propose a new No-reference Image Quality metric named WG-LAB, based on extraction of a set of features from distorted images in multiple domains (i.e., wavelet, and gradient domains) and multiple color channels (i.e., L, a, and b). Features are then exploited using Relevance Vector Machine Algorithm (RVM). The predictive performances of our method have been evaluated and compared to subjective judgments in terms of correlation, monotonicity and accuracy using the LIVE image database release 2 (LIVE II). The predictive performances obtained for our metric show that it has interesting features when compared to an array of existing full-reference (FR) and no-reference (NR) metrics on LIVE database.