Error Resilience In Scalable Video Coding Extension Of H.264/AVC Standard

Limited bandwidth recourses lead to a number of challenges especially for video communication over Internet Protocol (IP) networks. Scalable Video Coding (SVC) utilizes predictive coding to achieve high compression to overcome this limitation. However, predictive coding also makes SVC bit-streams vulnerable to transmission errors as the highly compressed videos suffer from both spatial and temporal error propagation resulting in the quality degradations. Consequently, error resilience techniques are utilized to combat transmission errors in SVC-based video communication, which forms the core issue discussed in this thesis.This thesis addresses error resilience techniques applicable for SVC-based video communication, explores various schemes to overcome error propagations which degrades the quality of the reconstructed videos at the receiver, and enhances the error robustness in the packet loss environment for heterogeneous users.In the first contribution, we address impacts of transmission error in quality degradation for hierarchical prediction structure in which the gap between two key pictures may be large. The conventional redundant-key-picture error resilience with picture-copy error concealment method may not be efficient due to copy with large distance to the lost picture. Exploiting strong correlation of the motion information between two consecutive key pictures, a combination of redundant-key-picture error resilient coding with motion-copy error concealment is proposed at the decoder. Simulations result show that proposed combination outperform existing solution, especially for unknown channel loss rates scenarios.In the second contribution, we formulate mode decision of loss-aware rate distortion optimization (LA-RDO) with Lagrange multiplier selection in SVC scenarios. The existing scheme use same Lagrange multiplier for each layers in LARDO-based SVC encoding. Exploiting quantization correlations between two consecutive layers, multi- layer Lagrange multipliers for optimal mode decision are proposed by joint optimization of base and enhancement layers. Simulation results show that proposed algorithm provides significant PSNR gain and bit rate saving with quality and spatial scalabilities for IPPP coding as well as for hierarchical B picture coding.The third contribution of the thesis addresses use of inter-layer residual prediction on trade-offs between concealment of lost macroblocks and reduction of visual artifacts problem in Extended Spatial Scalability (ESS). Whereas, the existing schemes focuses only to reduce artifact, and may not be efficient for transmission error conditions. A modified scheme is proposed for efficient use of residual prediction's trade-offs by exploiting homogenous characteristics in video objects. Simulation results showed that proposed schemes outperform existing scheme for subjective and objective qualities specifically for video communication over wireless channels.Finally, the forth contribution of this thesis address reduction of interpolation complexity in motion compensation for the mode derivation in error prone environment. The existing approach uses block-merging scheme to reduce complexity and to improve efficiency. However, to conceal lost picture less computations are required in motion compensated prediction for large mode partitions as compared to small partitions. We propose modified macroblock mode prediction with block merging at the ESS enhancement layer exploiting features in homogenous characteristics in video objects. Experimental results reveal that proposed algorithm achieves significant complexity reduction with increases in the number of larger mode partitions and improves error resilience to error prone transmission conditions.