Defining a no-reference image quality assessment by means of the self-affine analysis

In this paper we propose a novel Blind Image Quality Assessment via Self-Affine Analysis (BIQSAA) method by considering the wavelet transform as a linear operation that decomposes a complex signal into elementary blocks at different scales or resolutions. BIQSAA decomposes a distorted image into a set of wavelet planes omega(lambda, phi) of different spatial frequencies lambda and spatial orientations phi, and it transforms these wavelet planes into one-dimension vector omega using a Hilbert scanning. From the vector omega there were obtained their wavelet coefficient fluctuations estimated by the inverse of the Hurst exponent in decibels, whose scaling-law or fractal behavior was obtained by applying Fractal Geometry or Self-Affine Analysis. The scaling exponents calculated for the coefficient fluctuation behavior of Image Lena at 24bpp, at 1.375bpp, and at 0.50bpp were H-24bpp = 0.0395, H-1.375bpp = 0.0551, and H-0.50bpp = 0.0612, respectively. Our experiments show that BIQSAA algorithm improves in 14.36% the Human Visual System correlation, respect to the four state-of-the-art No-Reference Image Quality Assessments.

Automated summary

This paper introduces a novel Blind Image Quality Assessment method, BIQSAA, which considers self-affine analysis in wavelet transformation. The method decomposes distorted images into wavelet planes of different spatial frequencies and orientations, transforming them into one-dimensional vectors for analysis of wavelet coefficient fluctuations. Experiments show that BIQSAA algorithm improves Human Visual System correlation by 14.36% compared to state-of-the-art No-Reference Image Quality Assessments.