No-reference and Reduced Reference Video Quality Metrics: New Contributions

Digital video communication has evolved tremendouslyin the past few years, experiencing significantadvances in compression and transmission techniques. To quantify the performance of a video system, it isimportant to measure the quality of the video. Sincehumans are the ultimate receivers of a video signal, quality metrics must take into account the propertiesof the human visual system. So far, most of themetrics that have been proposed require access to theoriginal video, what makes them unsuitable forreal-time applications. We investigate how toestimate video quality in real-time applicationsusing no-reference and reduced reference metrics. Forthis, we study the visibility, annoyance, andrelative importance of different types of artifactsand how they combine to produce annoyance. The workuses synthetic artifacts that are simpler, purer, andeasier to describe, allowing a high degree of controlwith respect to the amplitude, distribution, andmixture of different types of artifacts. We presentmetrics for estimating the strength of four types ofartifacts. The outputs of the best artifact metricsare used to build a combination model for overallannoyance.