The Research On Rate Allocation Algorithm In Distributed Video Coding

With the wide development of mobile and wireless networks, many new applications arise, such as wireless video surveillances, sensor networks and mobile video phones, these applications require light encoding complexity. However, due to heavy encoding complexity, it's not so easy for conventional video coding such as MPEG or H.264 to be applied to these applications. Under this background, a new coding paradigm called distributed video coding (DVC) has attracted much more attention, Unlike the conventional video coding methods, DVC has low comlexity at the encoder, low power and good error-resilience, and etc, it can provide better solutions for the above metioned emerging applications.One of the most difficult tasks in DVC is rate allocation, this is mainly because of DVC frames are encoded separately and decoded jointly, the encoder often has no access to the motion compensation information at the decoder. It's difficult to accurately estimate how many parity bits to send for each frame without exploiting those information. In this case, most existing DVC solutions assume there is a feedback channel (FC) beteween the encoder and decoder. However, in many video coding applications, such as low-complexity video acquisition-and-storage devices, the FC is not allowed. Moreover, the FC will lead to latency, and can only be used in online applications, it is not suitable for those applications. So it's indeed extremely meaningful for us to research rate allocation algorithms without FC and having lower compexity at the encoder.This paper firstly reviews the theoretic background and the research development in DVC, and introduces two classical DVC architecture. Then, the paper analyses the performance of existing schemes with or without FC in DVC. Moreover, this paper mainly researches on one of the rate allocation schemes without FC in DVC, In this scheme, the rate for a particular bitplane has to be estimated at the encoder, this needs to determine the relationship between probability of error and parity groups, however,there isn't one but many parity groups for each probability of error, so it's very difficult to estimate the rate accurately at the encoder. In this paper, starting with the relationship between error propability of a bitplane and parity groups, we treat rate allocation in DVC without FC as a classification problem, and then propose a new scheme to estimate rate based on BP-neural network. In order to achieve more accurate rate allocation, besides of error propability, other factors which have influence on parity groups such as error rate, burst error and error distance, are all considered as feature vectors of the BP-neural network. Experimental results on test video sequences show, the rate allocation algorithm proposed in this paper can obtain more accurate results than a previous algorithm, the improvement for a bitplane can be as large as 63.58%.