HVS-metric-based performance analysis of image denoising algorithms

Image filtering is applied in numerous applications. Although most images after processing are subject to visualization and analysis by humans, MSE and PSNR metrics that do not adequately characterize image visual quality are still basically used in filter design and performance comparisons. This paper presents and generalizes some recent results obtained during our studies on usefulness of novel quality metrics based on human visual system (HVS). Basic attention is paid to DCT-based filters although some other denoising techniques are considered as well. It is demonstrated that the same DCT-based filter (with a tunable parameter) provides optimal values of PSNR and HVS-metrics for different values of tunable parameter. Local analysis of visual quality metrics is carried out indicating directions of the future research.