Anil Fernando
Video Coding technologies have evolved tremendously over past decades in line with the rapidly increasing demand over vastly expanding application domains. The research on video coding has been traditionally dominated by the work on MPEG and ITU-T H.26x standardisations based systems. It is known that in these mainstream technologies the video encoders are far more complex (by approximately 5 to 10 times) than the decoder structure. This architecture was motivated by many of the conventional one-to-many type video applications including broadcasting (DVB), video streaming etc where the decoder cost need to maintained considerably low for the benefit of large numbers of viewers compared to the limited number of content providers. However, more recently, this architecture is challenged by new consumer applications where the cost of encoder is a prime concern due to the necessity of vast deployments of video sensors. Security surveillance systems, mobile video conferencing, monitoring of the disabled people and children, disaster zone monitoring are a few potential scenarios which are largely benefited by massive encoder deployments. Distributed Video Coding (DVC) is an emerging video coding technology designed with a modified complexity balance between the encoder and decoder in line with the necessities of these applications. The dramatically low complexity of the DVC encoder helps these solutions by: (i) reducing the production cost of the signal processors of the video sensors, (ii) reducing the requirement of digital memory and (iii) reducing the power consumption which is generally a scarce resource at remote sites.
The research on DVC is still in preliminary stages and considerable amount of effort is necessary before going through the standardisation process and commercial use. The currently available literatures have used a number of hypothetical models and assumptions some of which have not yet been assessed for practical viability. Key frame transmission algorithm, error and noise distribution estimation at the decoder, implementation of the reverse feedback channel using the dynamic error estimation and the necessary communication protocols are some of the major open areas for research in DVC. Side information generation is an area largely discussed in literature, yet further room for development.
In DVC, the shift of complexity balance is achieved by moving the major task of the exploitation of source correlations to achieve the compression into the decoder. This task involves the generation of a representation of a part of original sequence called side information. A sequence of ‘selected’ original frames is generally passed to the decoder over the channel using an intra frame coding scheme and are called ‘key-frames’. The frequency of key-frame transmission could vary on the DVC implementation strategy. The missing frames are estimated at the decoder using interpolation/extrapolation techniques or more complex and accurate motion prediction methods. It is assumed that the side information so generated is a form of the original sequence subjected to noise while transmission. The identification of the statistical distribution of this ‘noise’ is a part of the ongoing research activity. The side information used for processing with the parity information sequence transmitted over the channel by the encoder. At the encoder this parity bits are generated by passing the original video sequence through a set of shift registers and logic gates. Further this parity bit sequence is generally subjected to puncturing, of which the rate determines the channel bandwidth requirement and it varies on the implementation strategy for a given quality of image reconstruction at the decoder.
The number of bits transmitted over the channel to represent each pixel (bpp) and the closeness of the reconstructed image at the decoder to the original frame held back at the encoder (PSNR) are the common measures of the goodness of the video codec implementation. The theoretical base and the guidelines for Distributed Source Coding were set by Slepian-Wolf and the current work in this field is based on the work by Wyner-Ziv.
In this tutorial we will discuss the motivations behind the DVC design, current DVC codec architecture and possible modifications to enhance the performance, the hypothetical models, assumptions used in the current design, design criteria for possible practical solutions and some of the potential application domains.
W.A.C. Fernando (SMIEEE) leads the Video Codec group in University of Surrey, UK. He has been working in video coding since 1998 and has published more than 155 international refereed journal, two book chapters and proceeding papers in this area. Furthermore, he has published more than 45 international refereed journal and conference papers in DVC. He is also attached to the VISNET European project which covers lots of DVC activities as the leading institute. He has also lots of research collaborations in DVC within Europe and North America and Asia.
He is a member of the editorial board of the international journal of multimedia tools and applications. Recently, he has been nominated as the guest editor for two special issues on DVC (IET journal of Image Processing and journal of multimedia tools and applications). Furthermore, he has been working as a referee for IEEE Transactions on Circuits and Systems for Video Technology, IEEE Transactions on Communications, Mobile Computing, Communications Letters, IEE proceedings of communications, IEE proceedings of Vision and Computing, IEE Electronic Letters, Journal of Communications Networking, Electronics and Telecommunications Research journal, SPIE journals, etc., and many conference proceedings (VTC, ICC, ISCAS, ICIP, SPIE, ITC, etc.,).