A generic method to increase the prediction accuracy of visual quality metrics

Abstract

A new simple, effective and generic method is presented that allows to increase the prediction accuracy of visual quality metrics. This method is based on the observation, that the correlation between subjective results and objective metrics is very high for one special sequence or image coded at several bit rates with the same codec. This can be used to decrease the absolute error between the predicted visual quality and the actual visual quality and therefore increase the overall correlation between prediction results and results from subjective tests. Using the described method PSNR is extended to PSNR +. Comparing this new metric PSNR + to two popular image quality metrics shows that the prediction accuracy can be increased significantly. Index Terms — Video quality metric, visual quality, AVC/H.264, PSNR