Study of MPEG-2 coding performance based on a perceptual quality metric

Study of MPEG-2 Co ding Performance based on a PerceptualQuality MetricAndrea Basso* **_Ismail Dalgc*Fouad A. Tobagi* andChristian J. van den Branden Lambrecht**** Department of Electrical Engineering Stanford University** Telecommunication Laboratory Swiss Federal Institute of Tchnology*** Signal Processing Laboratory Swiss Federal Institute of TchnologyAbstractIn this paper some wel l known quality metrics suchas PSNR and the metric developed at Institute forTelecommunication Sciences (ITS) arereviewed.Their shortcomings in measuring quality of co-ded videocompared to subjective tests arepointedout. Then, a new video quality metric cal led Mo-ving Picture Quality Metric (MPQM) is presented.This metric models the human visual system andmatches correctly subjective evaluations. Compa-rative results in the case of constant bit rate (CBR)MPEG-2 codedsequences arepresented, showingthe superiority of MPQM over ITS and PSNR.1Intro ductionThe interest in multi-media applications - with astrong emphasis on video issues - is growing tre-mendously.Video assessment is a fundamentaland still not suciently explored asp ect of the cur-rent research on video co ding.Visual ob jectivemetrics that are coherent with quality as p erce-ived byhuman observers are b eginning to emergeonly recently.Furthermore very recently the con-cept of constant-quality video enco ding has b eenintro duced in [1] and further develop ed in [2].The motivation of this work is to determine theright kind of quality metric for devising a constant-qualityvariable bit rate (CQ-VBR) video enco dingscheme for MPEG-2 in view of an evaluation itsp erformances over ATM.In this pap er some well known quality metricsfor video such as the well known and widelyused Peak Signal to Noise Ratio (PSNR) and themetric develop ed at Institute for Telecommunica-its prop ertThis work was supp orted in part by NSF under grantNCR-9016032, and byPaci c Bell.tion Science (ITS) are reviewed. Then, a new vi-deo quality metric called Moving Picture QualityMetric (MPQM) is discussed. That metric is ba-sed on a mo del of the early stages of the human vi-sual system and it matches sub jectiveevaluationscorrectly. Results for the considered metrics areshown for constant bit rate (CBR) MPEG-2 co-ded sequences.The pap er is organized as follows. Sec. 2 over-views the literature on video quality assessmentand measurement. Sec. 3 illustrates the inadequ-acy of PSNR metric for video quality assessmentby means of a simple example. Sec.4 presentsthe ITS video quality metric and its p erformanceon MPEG-2 co ded video.Sec.5 discusses theMPQM. Some conclusive remarks end the pap er.2Stateoftheartinvideoquality assessment and me-asurement2.1First DevelopmentsVideo quality assessment plays a fundamental rolein the development of new and existing video co-ding algorithms.The interest for image qualityevaluation has b een strong since the sixties. Se-veral image quality metrics hae b een develop ed,such as the Strehl measure [3], based on the degra-dation to which an image is sub ject to, wheneverit passes di erent real optical systems compared tothe ideal case. Attempts with bivariate metrics ofimage qualityhave b een done in particular duringthe seventies. The reader is referred to [4] for areview.The mean square error has b een retained fory of b eing easily analytically tractable.Wilder [5] did a rather complete evaluation of dif-1