|M.Sc Student||Priziment Evgeniy|
|Subject||Modeling and Rate Control in Reversed Complexity Video|
|Department||Department of Electrical Engineering||Supervisor||Professor Emeritus David Malah|
|Full Thesis text|
Distributed Video Coding (DVC) is a novel coding scheme that employs principles of lossy source coding with side information at the decoder. In video applications, frame prediction, based on the adjacent frames, plays the role of side information. The distributed video coding framework enables to shift the computational load of motion estimation from the encoder to the decoder, resulting in reversed encoder-decoder complexity. This reversed complexity scheme could be appealing for applications in which the encoder is power and/or complexity constrained, such as wireless/cellular video, sensor networks and video surveillance. The operation of a DVC system is based on a joint distribution model of the source and side information.
In this work we first examine a family of stationary joint distribution models. As one of our findings, we propose to use the Gamma based model as an alternative to the widely adapted Laplace model, due to its superior performance. In addition, we suggest a new spatially adaptive model, which enables to follow the spatially varying joint statistics of the source and side information. Next, we present two methods, class-based and neighborhood-based, for estimation of the spatially varying model parameters. Further, we show how the model obtained in the pixel domain, can be used in the transform domain to facilitate frame spatial redundancy utilization.
Another problem addressed in this work is the rate control mechanism. Typically, in DVC systems, the rate control is performed by the decoder, through a feedback channel. The resulting latency might be unsuitable for real time applications. We propose an encoder-side rate control module. This module uses joint distribution parameters, which are fed to a rate distortion function, to obtain the required coding rate for a specified distortion level. Moreover, in order to exploit the spatially varying statistics, we suggest performing intra frame rate allocation, using the joint distribution together with the rate distortion function.
In addition, we studied a Block Truncation Coding (BTC) based DVC system, as a compromise between transform-domain and pixel-domain coding. We propose an enhanced reconstruction algorithm for the BTC frames, which utilizes the side information available at the decoder. We evaluated the developed models using standard test sequences. In addition, the models were tested on aerial video sequences, which are accompanied by metadata. This metadata describes the global motion in the sequence, which makes the joint statistics estimation process less complex and more reliable.