Analysis and Evaluation of Image Quality Metrics

Image Quality Assessment (IQA) is a very difficult task, yet highly important characteristic for evaluation of the image quality. Widely popular IQA techniques, belonging to objective fidelity, like Mean Square Error (MSE), Peak Signal to Noise Ratio (PSNR) or subjective fidelity which corresponds to the human visual system (HVS), like, Universal Quality Index (UQI), Structural SIMilarity (SSIM), Feature SIMilarity (FSIM), Feature SIMilarity for color images (FSIMc), Gradient Magnitude Similarity (GSM) have been discussed in this paper. Also quality measured on basis of degradation model and Noise Quality Measure (NQM) has been discussed. Experiments have been conducted on IVC database available online at http://www.irccyn.ec-nantes.fr/ivcdb/ and verified from the CSIQ database and LAR database available online at http://vision.okstate.edu/?loc=csiq and http://www.irccyn.ec-nantes.fr/similar to autrusse/Databases/LAR/. On the basis of the obtained values judgements about the image distortion and hence the optimum image quality metric has been decided. It has been found from all the experiments conducted that FSIM is the best metric for the JPEG, JPEG2000, blur and LAR whereas UQI failed to give better results for all except JPEG2000.