Reduced reference IQA based on structural dissimilarity

Distortion estimation in objective domain plays an important role during Image Quality Assessment (IQA). Reduced Reference IQA requires only partial information about the reference image which can be transmitted through an ancillary data channel and is useful where complete image cannot be accessed (as reference for IQA). In this paper, IQA is performed based on a reduced reference approach with minimal complexity. The proposed evaluation methodology involves decomposition of images into sub-bands using a suitable wavelet family for features extraction. These extracted features are then used to obtain a `structural dissimilarity measure' sensitive towards structural distortions. The computational complexity during estimation of features is reduced in this method by using combinations of integral image and gradient magnitude approaches. The aim of this approach is to analyse different distortions with a single reduced reference metric and maintaining its coherence with the Human Visual System (HVS). This has been justified in simulations by the healthy correlation of IQA between proposed reduced reference metric and human opinion.