<XML><RECORDS><RECORD><REFERENCE_TYPE>31</REFERENCE_TYPE><REFNUM>8599</REFNUM><AUTHORS><AUTHOR>Gunawan,I.P.</AUTHOR></AUTHORS><YEAR>2006</YEAR><TITLE>Reduced-reference Impairment Metrics for Digitally Compressed Video</TITLE><PLACE_PUBLISHED>PhD Thesis, Dept. of Electronic Systems Engineering, University of Essex, U.K.</PLACE_PUBLISHED><PUBLISHER>N/A</PUBLISHER><PAGES>xix,213</PAGES><LABEL>Gunawan:2006:8599</LABEL><ABSTRACT>This thesis presents an objective picture quality measurement tool for digitally compressed video in the absence of full-reference information. It focuses on the development of a reduced- reference video quality model, and particularly on the investigation of overhead data that can improve the prediction ability of the model. A reduced-reference method, based on global characteristics of picture, is developed using histograms of pixel amplitudes. Observations show that the histograms of digital images compressed at different bit rates exhibit some distinguishable patterns. The histograms are compared using dissimilarity metrics such as Kolmogorov-Smirnoff or ?2-test. Another reduced-reference method based on local characteristics of the picture is also presented. It uses a single tool to quantify multiple distortions such as blocking and blurring. A single analysis method based on examination of certain frequency components is developed. The reduced-reference data are constructed as matrix based on local harmonic strength (LHS) features of a picture. A discriminative analysis of the LHS matrices from the degraded and the original pictures results in two quality parameters, namely harmonic gain and harmonic loss that correspond with the blocking and blurring artifacts, respectively. The LHS-based overall quality metric is derived by combining the harmonic gain and loss information. This thesis also investigates the impact of motion in a video sequence on the LHS-based quality parameters and the way they should be weighted and combined to give the best objective video quality metric. It is also found that the reduced-reference data based on the LHS matrix is flexible; i.e., we can reduce the size of its elements without compromising the performance of the model. In addition, a development into a simple no-reference model is also considered, since such a model can be regarded as reduced-reference with zero bandwidth overhead. The no-reference model developed in this thesis is based on a modification of the PSNR/MSE without resorting to the human visual system. The rationale of the method is to retain the computational simplicity of the PSNR/MSE whilst removing the requirement of using the original picture as reference. In place of the original picture, some post-processed versions of the degraded picture are used as reference ‘proxy’. Experimental results show that the proposed models presented in this thesis produce good correlations with publicly available subjective data. The performance of the methods presented are also comparable to those of full-reference models in the literature.</ABSTRACT><URL>http://serlib0.essex.ac.uk/search/a?SEARCH=Gunawan&sortdropdown=-</URL></RECORD></RECORDS></XML>