Reduced reference image quality assessment based on tetrolet statistics
Abdelkaher Ait Abdelouahad, Mohammed El Hassouni, Hocine Cherifi · International journal of imaging and robotics · 2018
In this paper, we present a reduced reference image quality assessment (RRIQA) measure, in the Tetrolet transform domain, using natural scene statistics (NSS) approach. The tetrolet transform provides a convenient way to capture local geometric structures. We propose the Gaussian Scale Mixture (GSM) model to characterize the dependencies between tetrolet coefficients. First, the tetrolet transform is applied to the reference and distorted images. Then, a set of statistical features are extracted from tetrolet representations of both images. Finally, a comparison between those features is conducted to quantify the visual degradation. Four datasets were used to evaluate the performances of the proposed measure (LIVE, Cornell- VCL A57, IVC and TID2008), significant improvements were obtained for several distortions. Also, the proposed measure does not depend on the type of distortions or require any training. As a consequence it is a general purpose RRIQA method.