Low-complexity computation of visual information fidelity in the discrete wavelet domain

Abstract

The visual information fidelity (VIF) index is an objective quality metric that gives very accurate image similarity scores, but at the cost of very high computational complexity. In this paper, a method is presented for calculating VIF in the discrete wavelet domain using the Haar wavelet. The proposed method exploits scalar Gaussian Scale Mixture (GSM) instead of vector GSM for calculating the prediction scores. The complexity of the proposed method is assessed for five different popular image sizes and compared to other methods based on a C/C++ implementation of the algorithms. Experimental results show that the proposed method can compute the visual quality score with less than 30% of the computational complexity of the well-known SSIM index, with greater accuracy than that achieved by the original VIF index method (at about 5% of its computational complexity).