Mobile Video Communications Using a Wyner-ZivTranscoderEduardo Peixotoa, Ricardo L. de Queirozaand Debargha MukherjeebaDepartamento de Engenharia Eletrica, Universidade de Brasilia, Brasilia, Brazil;bHewlett Packard Laboratories, Palo Alto, CA, USAE-mail: [email protected], [email protected], [email protected] mobile-to-mobile video communications, both the transmitting and receiving ends may not have the necessarycomputing power to perform complex video compression and decompression tasks. Traditional video codecstypically have highly complex encoders and less complex decoders. However, Wyner-Ziv (WZ) coding allows fora low complexity encoder at the price of a more complex decoder. We propose a video communication systemwhere the transmitter uses a WZ (reversed complexity) coder, while the receiver uses a traditional decoder,hence minimizing complexity at both ends. For that to work we propose to insert a transcoder in the network toconvert the video stream. We present an efficient transcoder from a simple W Z approach to H.263. Our approachsaves a large amount of the computation by reusing the motion estimation performed at the WZ decoder stage,among other things. Results are presented to demonstrate the transcoder performance.Keywords: Wyner-Ziv, video transcoding, mobile video, H.263.1. INTRODUCTIONMobile-to-mobile video communications is a highly desirable goal in personal communications. Nevertheless, therequirements to have low complexity at both encoder and decoder sides have been prohibitive. These requirementsdo not allow traditional codecs to work in their best modes, so that neither the rate nor the quality are at theirpeak. The traditional codecs are more complex at the encoder side, where most of the computation is doneto decide the best way to encode each macroblock. The decoders, however, are less complex, just performingentropy decoding, motion compensation and inverse transforms. Reverse complexity video codecs are a newparadigm, where the encoder has low complexity and the decoder exploits the video statistics, thus being morecomplex. Neither traditional codecs nor reverse complexity codecs are suitable to video communications wherelow complexity is required at both ends. To solve this problem, we propose a transcoder from a reverse complexitycodec to a traditional codec, thus achieving the best of each codec. Such a transcoder has already been mentionedin the literature,1however, to our knowledge, it has never been implemented.The traditional video compression solutions are based on the hybrid architecture of block-based motioncompensated prediction and on a spatial transform followed by entropy coding of the residue and related infor-mation.2, 3The goal of these tools is the reduction of spatial and temporal redundancy. In this architecture, theencoder has a higher complexity than the decoder. Typically, the encoder is many times more complex than thedecoder, mainly due to motion estimation.2–4These codecs are suited to applications where the video contentis encoded once and decoded multiple times, e. g. home DVDs, or decoded by many devices, e.g. broadcasting.Distributed video coding (DVC) is based on reverse complexity codecs, where the encoder is less complexthan the decoder.1DVC allows us to explore the video statistics at the decoder side, theoretically achieving thesame rate distortion of a traditional coder (which exploits the video statistics at the encoder). These codecs arebased on the Slepian-Wolf5and Wyner-Ziv6theorems. In a Wyner-Ziv (WZ) codec, the encoder is simpler, ata price of a more complex decoder. They are suitable to applications where the enco ding has to be performedwithin a low resources enviroment, but the decoding can be made at a more powerful machine or offline.While DVC fits well an encoder for a mobile camera, it just shifts the need of computational resources to thedecoder. The goal of low encoding complexity and low decoding complexity, within a single scheme, and withoutgiving up quality, seems unreacheable. In this paper, we propose a transcoder from a WZ scheme to a traditionalFigure 1. The transcoder within the proposed framework.hybrid DPCM/DCT scheme. In this scenario, a mobile camera phone performs WZ encoding, transmitting thedata to a server. The server acts as a transcoder, changing the video sequence to another format, which requiresa less complex decoder. The transcoded data is sent to another mobile device, the real addressee of the videocontent. This scheme is shown in Fig. 1.2. THE CODECS USEDTo achieve our goal of low complexity at both ends, we need two codecs: one with a low complexity encoder andthe other one with a low complexity decoder. Traditional codecs usually have low complexity decoders, sincethey explore the video statistics at the encoder. The codec chosen to implement our framework is the H.263.7Itwas chosen because it is an efficient ITU-T standard and it is significantly less complex than the newest, state ofthe art video codec H.264/AVC,8therefore reducing the total complexity of the transcoder and of the decoder.WZ coding has been the focus of many studies in recent years,9,10,11.12The WZ codec used in this pap erwas recently proposed.9It is a simple pixel-domain WZ codec, whose architecture is shown in Fig. 2.Figure 2. Wyner-Ziv encoder and decoder framework.At the encoder, some frames (called key frames) are encoded with a regular intra-frame encoder (in thiscase, H.263 INTRA). The other frames (called WZ frames) are encoded as follows: the difference between thecurrent frame, X, and the encoder reference frame, Xer, is fed to the WZ encoder. Xeris taken as the previousreconstructed key frame, which is available both to the encoder and decoder, in order to avoid drifting errors.The pixels of the residual frame R = X − Xerare quantized using an uniform scalar quantizer, generating RQ,which is then reordered and encoded by a low-density-parity-check accumulated (LDPCA) code.13The WZencoder may have a group of pictures (GOP)2length of 2 (1 key frame followed by 1 WZ frame), 3 (1 key framefollowed by 2 WZ frames), or even more. More WZ frames may be encoded b etween each key frame, decreasingthe encoder complexity. However, the correlation of the side information decreases as the GOP length increases,thus reducing the quality of the encoded sequence.The decoder generates the side information (SI) Y using motion-compensated interpolation. The SI generationprocess is
View Full Document
Unlocking...