A new method for no-reference image Blockiness measurement

As video traffic over Internet and service provider networks peaks today, the need for image quality measurement is obvious. Such measurements need to be done real-time, and that requires “No-Reference” measurements, i.e., measurements that do not use the uncompressed (raw) images. Now, great majority of video streams are digital. Digital video streams employ compressed streams, and they are DCT-based. In such streams, quality degradation results from reduced bitrates, and since DCT algorithm is block-based, quality degradation manifests itself as “Blockiness”. In this work, we present a blockiness measurement method (called RED) that gives results closer to human perception than the literature. One of the main contributions of RED is the analytic relationship it establishes between its automatically produced blockiness metric and the scores obtained from human testers. RED optimizes the parameters of this relationship through “Regression”. Another unique feature of RED compared to the literature is that it skips over some DCT blocks with the help of “Edge Detection”.