UBC Theses and Dissertations
Low bit rate DCT-based video coding in error-prone environments Côté, Guy
Video compression, which is a necessary process for video communication over low bit rate networks, removes spatial and temporal redundancies contained in video sequences. However, when transmitted over error prone networks, compressed video streams are very sensitive to bit errors and packet losses, which lead to inevitable spatial and temporal error propagation. This thesis presents efficient video encoding methods for transmission over error prone networks. The resulting video communication systems exploit the video input statistics, the channel condition information, and the encoder's knowledge of the video decoder's error handling capabilities to optimize the encoded bit stream for its transmission over error prone networks. We first propose a rate-distortion optimized synchronization marker placement algorithm which can effectively determine the frequency and location of synchronization markers, subject to a known decoder concealment method. Next, we present a semi-fixed length coding method for the efficient entropy coding of motion vectors. With the use of unequal error protection techniques, the proposed method provides superior bit stream and spatial synchronization, compared to variable length coding methods. A new algorithm for temporal error resilience employing optimal intra block updating is then presented. Using a new distortion measure that takes into account the effects of temporal error propagation, the proposed method chooses the amount of non-predictive coding necessary to achieve temporal synchronization. We combine the proposed methods and evaluate the performance of two resulting video communication systems in error prone environments, using packet loss and bit error models for the Internet and mobile networks, respectively.
Item Citations and Data