No-reference image quality assessment based on local binary patterns

Abstract

This paper presents the new algorithm for no-reference image quality assessment (NRQ LBP). This algorithm does not need a priori information about possible types of image distortions before assessment. No transformation to another coordinate frame (DCT, wavelet, etc.) is required, distinguishing it from prior no reference quality assessment approaches. NRQ LBP is based on machine learning and uses extremely randomized trees method for mapping quality features with subject quality score (DMOS). Quality features are bins of a histogram of local binary patterns calculated for neighborhood radiuses 1, 2, 3 pixels. Comparative experimental results a given for modern image quality assessment algorithms (PSNR, SSIM, MS-SSIM, LBIQ, GRNN, BRISQUE, NRLBPS). Images from standard LIVE database are used as training and testing datasets. Spearman correlation coefficient, Pearson correlation coefficient and RMSE are used to determine the accuracy of compared algorithms. Performance results shows that proposed algorithm is highly competitive with tested algorithms and moreover it has very low computational complexity, making it well suited for real time applications.