Multiple description image coding based on lapped orthogonal transforms

This dissertation considers the use of multiple description coding. (MDC) for image transmission in communication systems where long burst errors and sometimes complete channel failures are inevitable. The basic idea behind MDC is to introduce a desired amount of correlation between multiple descriptions. Although this correlation will reduce the coding efficiency compared to the conventional single description (SD) coder designed to maximize the coding efficiency solely, it enables the decoder to reconstruct an acceptable signal from any description, because each description is correlated with other descriptions and carries sufficient information about the input image signal. The design objective of a MDC system is to minimize the redundancy required to achieve a given distortion criterion.; This thesis proposes to realize the MDC based on lapped orthogonal transforms (LOTs), by providing non-hierarchical image decomposition at the encoder and reconstruction at the decoder. The encoder generates multiple coded descriptions by using conventional overlapping block transforms, and then splitting the resulting coefficient blocks into multiple (four) descriptions. Within this coding framework, the two important problems to be solved are: (i) how to recover an image from an incomplete set of descriptions, and (ii) how to design the LOT basis to minimize the redundancy required to achieve a given reconstruction distortion. A novel image reconstruction technique, called maximally smooth recovery (MSR) method, is developed for the decoder. This method makes use of the constraints between adjacent LOT coefficient blocks and the smoothness property of common image signals, and facilitates image recovery from only available descriptions or from incomplete sets of LOT coefficient blocks. Furthermore, a new basis design method which controls the trade-off between the coding efficiency and the error resilience, is developed by adjusting a weighting between the coding gain and the reconstruction gain. Simulation results show that the newly designed bases achieve significantly better trade-off between the coding efficiency and the error resilience than previously designed bases. In other words, they can provide better reconstruction quality with the same amount of bitrates, or vice verse.