QUASI-BLIND ON LINE VIDEO QUALITY TRACKING BASED ON POLAR EDGE COHERENCE V. Baroncini (*) , L. Capodiferro (*) , E. D. Di Claudio (**) , G. Jacovitti (**) , F. Mangiatordi (*) , G. Ridolfi (***) (*)Fondazione “Ugo Bordoni” Via Baldassarre Castiglione 59, I-00142, Rome, Italy web: http:\\www.fub.it (**)INFOCOM Dpt., University of Rome “La Sapienza” Via Eudossiana 18, I-00184, Rome, Italy web: http:\\infocom.uniroma1.it (***) RAI-Direzione Strategie Tecnologiche web: http:\\www.rai.it ABSTRACT A quality assessment method for video sequences is presented as an evolution and extension of a recently introduced technique. The method is quas- blind, since it compares quality indices independently calculated for the examined video sequence and for a reference version of the same sequence. The employed statistics are based on edge and moving edge polar coherence calculated with pairs of circular harmonic functions. Key Words: video quality assessment, edge coherence, Laguerre Gauss. 1. INTRODUCTION Reduced Reference (RR) approaches make use of auxiliary low-capacity channels to share selected features of the original images with the user in telecommunication applications. These features are compared with the corresponding ones of the observed image for video quality assessment. Alternatively, some (RR) approaches measure specific distortions the video sequence undergoes in specific cases. Examples of RR techniques are the ones presented in [1], [2], [3] and [4]. In this contribution we present a quasi-blind video quality assessment method, which requires the transmission of a single number per frame or group of frames, so reducing the practical extra information to negligible overheads. The method is based on measuring the quality of reproduction of edges, called polar edge coherence. This metric is derived by a local image decomposition with the so-called Laguerre-Gauss (GL) family of angular harmonic functions. It is not oriented to specific distortions such as, for example, JPEG blocking, but is sensitive to different impairment causes. In particular, the measured coherence tends to be disrupted on the average by image compression techniques [5]. It is shown here how the repetition of said measurements frame by frame, allowed by its parsimonious nature, can be exploited. This feature is especially interesting for instance for the quality evaluation in mobile communications networks where video streams are subject to a large variety of distortions caused by compression, channel effects, concealment, post-processing etc. In those cases the polar edge coherence measurement enables for on-line quality monitoring and dynamic resource allocation to meet specific QoS requirements. In fact, it is shown how frame by frame tracking of the proposed indices along the sequence allows to take into account the dynamic effects of transient quality losses. Proceeding further on, the measurement of polar edge coherence is extended to video sequences by measuring the coherence of moving edge through adjacent frames. To this purpose, an interframe edge coherence metric is defined using a space-time local image decomposition with the Laguerre-Gauss (LG) family. Preliminary experiments yields interesting correlation with perceptual findings. . 2. THE LAGUERRE GAUSS PLEXUS The measurement of the edge coherence is performed through a local image representation made by the Laguerre Gauss expansion. With reference to the polar coordinates 2 2 x y r + = , y x arctg ϕ              = , the image (,) Ir γ referred to the generic analysis point can be expanded as follows: ( ) , , (,) ,; nk nk n k Ir y r γ γσ ∞ ∞ =−∞ =−∞ = ∑ ∑ L