Enhancement of Wyner-Ziv video coding for wireless channels

Wyner-Ziv video coding (WZVC) is a fast emerging video coding technique for wireless video sensor networks. WZVC moves the complexity from the encoder (sensor) to the decoder (receiver). This thesis proposes few enhancements to solve challenging problems in WZVC, namely 1) handling impairments of a fading wireless channel in WZVC environment, 2) investigating rate penalty of WZVC in wireless fading channels, 3) adaptive encoder rate control using inter-frame cross-correlation properties, and 4) better side information estimation with a bit-based noise variance computation technique. In case (1), the decoder metric values in WZVC are calculated with respect to the correlation noise, channel noise, and fading. Multiple input multiple output (MIMO) diversity scheme is studied with WZVC for improving the reconstructed video output. Simulation results show that the average peak signal to noise ratio (PSNR) of Foreman video is improved by ≈ 3 dB with the configuration of 2I4O compared to single input single output (SISO) channel at SNR = 2 dB. In case (2), expressions for the rate penalty are analytically derived under different wireless channel conditions. WZVC scheme with receiver diversity (WZVC-RD) is proposed and also demonstrated that it alleviates the channel-induced rate penalty. Simulation results show that with adequate diversity, the channel induced rate penalty can be almost completely eliminated, i.e., the average rate penalty is reduced to be less than 20 Kbps in WZVC-RD for Foreman video at SNR = 2 dB. In case (3), theoretical rate-distortion behavior of conditional decoding is compared with that of the practical WZVC. An encoder rate control algorithm is proposed with respect to the cross-correlation threshold (CCTH) at the decoder, where the definition of the threshold is derived based on cross-correlation statistics of the key frames. Additionally, an adaptive threshold algorithm (ATHA) is proposed to improve the PSNR versus rate relationship. In case (4), correlation noise variance estimation is proposed with respect to the bit pattern of each pixel; it is named as “bit-based noise variance”. PSNR improvement of ≈ 4 dB is observed for the Foreman video frames with higher correlation noise.