Structural similarity weighting for image quality assessment

Recently, there has been a trend of investigating weighting/pooling strategies in the research of image quality assessment (IQA). The saliency maps, information content maps and other weighting strategies were reportedly to be able to amend performance of IQA metrics to a sizable margin. In this work, we will show that local structural similarity is itself an effective yet simple weighting scheme leading to substantial performance improvement of IQA. More specifically, we propose a Structural similarity Weighted SSIM (SW-SSIM) metric by locally weighting the SSIM map with local structural similarities computed using SSIM itself. Experimental results on LIVE database confirm the performance of SW-SSIM as compared to some major weighting/pooling type of IQA methods, such as MS-SSIM, WSSIM and IW-SSIM. We would like to emphasize that our SW-SSIM is merely a straightforward realization of a more general framework of locally weighting IQA metric using itself as similarity measures.