An Adequate Objective Performance Evaluation of Dirac Video Codec Compared to H.264/AVC André F. de Araujo Communications Department FEEC/UNICAMP Campinas, Brazil aaraujo@decom.fee.unicamp.br Yuzo Iano Communications Department FEEC/UNICAMP Campinas, Brazil yuzo@decom.fee.unicamp.br Rangel Arthur School of technology UNICAMP Limeira, Brazil rangel@ft.unicamp.br Abstract— Dirac, a relevant new open wavelet-based video codec, has been released recently. The main contribution of this paper is an adequate performance evaluation of this codec compared to the well-known H.264/AVC. This work differs significantly from previous ones which present unfair and inconclusive comparisons. Eight videos comprising three different resolutions with various degrees and kinds of motion are used to perform the simulations, as well as appropriate tuning of coding parameters for both codecs. In every case, H.264’s objective visual quality results outperform Dirac ones, but some evidences indicate Dirac might provide better subjective quality. Keywords-Dirac video codec; H.264; Video Coding; Vide Compression; Wavelet-based video coding. I. INTRODUCTION Video coding is essential to enable video transmission and storage in modern communications systems. Most important video codecs, such as MPEG-based ones (and then H.264/AVC), employ the Discrete Cosine Transform (DCT) – or a similar transform based on it – as the transform responsible for eliminating spatial redundancy. The Discrete Wavelet Transform (DWT) has gained importance recently, as it presents important advantages when compared to DCT, such as the possibility of working with the whole picture, instead of the need to partition into squared blocks. Moreover, subband decomposition results in more flexibility in terms of scalability in resolution and distortion. Dirac is a recent open (royalty-free) video codec developed by the BBC R&D. Its architecture is similar to hybrid MPEG- based ones (as in [1]), except that the DWT is used for eliminating spatial redundancy. It was conceived to provide visual quality at H.264/AVC level. Dirac creators explicitly argue that the codec was not designed to optimize objective visual quality measurements [2]. However, objective evaluations still provide important indices of the resulting quality and are able to indicate the general degree to which the compression is achieved. Some works [3]-[6] have already attempted to provide an objective evaluation of Dirac against H.264, but these analyzes do not take into account important parameters so that the codecs can be compared adequately. These are then inconclusive. In this work, Dirac is considered comparatively with regard to H.264/AVC, aiming to provide objective visual quality results for both codecs, so that a fair comparison is obtained. Initially, an overview of the main features of Dirac is outlined. Experimental results are derived using eight videos with three different resolutions and various degrees and kinds of motion with all relevant parameters adequately tuned. II. OVERVIEW OF DIRAC VIDEO CODEC Dirac is an open video codec developed by the BBC. It provides great compression efficiency (it is intended to compete with H.264/AVC) with a simple architecture, based on a small number of core tools. Its overall design is similar to a conventional hybrid motion-compensated codec, except that the function usually performed by a block transform in most standards (DCT, in most cases) is instead performed by the wavelet transform. Fig. 1 presents the overall architecture of the codec. The encoder consists in four major modules, supporting both frames and fields: Transform and Quantization, Motion Estimation, Motion Compensation and Entropy Coding. Figure 1. Overall architecture of Dirac Video Codec [7]. This work was supported by CAPES, RH-TVD/CAPES and FAEPEX/UNICAMP