Multiple description video coding with side information

contact: O. Crave, C. Guillemot, B. Pesquet-Popescu

Context and Goal

Multiple description coding (MDC) has been introduced as a generalization of source coding subject to a fidelity criterion for communication systems that use diversity to overcome channel impairments. MDC is an interesting tool for robust communication over lossy networks such as the Internet, peer-to-peer, diversity wireless networks and sensor networks. MDC with side information at the receiver is particularly relevant for robust transmission in sensor networks where correlated data is being transmitted to a common receiver, as well as for robust video compression where the encoder's low complexity is a requirement. Here, we focus on the design of a practical multiple description video coding scheme with side information at the receiver. It builds upon both MDC principles and Slepian-Wolf (SW) coding principles.

Approach

The idea is illustrated in Fig. 1, using two descriptions as an example. At the transmitter, a subset of frames, regularly spaced in the sequence, is compressed by a key frame video coder in intra mode. The frames between the key frames are Wyner-Ziv coded. Each coder generates two descriptions using multiple description scalar quantization (MDSQ). One description of key frames is combined with a description of Wyner-Ziv frames to create a new description for the entire input signal. Both descriptions D1 and D2 are sent on a loss-prone channel to the decoder. At the decoder side, if only one description is received, the low quality of the side information generated from the key frames should be sufficient to decode the Wyner-Ziv frames, and the entire sequence could be recovered with an acceptable quality. When both descriptions are received, the descriptions for the key frames and the Wyner-Ziv frames are jointly decoded and should lead to a better quality than at the side decoders for every frame of the sequence.

Fig. 1 - Two-description video codec with side information.

The MDSQ introduces redundancy or correlation in the transmitted streams in order to take advantage of the network path diversity. In the case of the Wyner-Ziv frames, the resulting sequences of indexes are SW encoded, that is separately encoded and jointly decoded. While the first step (MDSQ) plays the role of a channel code, the second one (SW coding) plays the role of a source code, compressing the sequences of quantized indexes.

Experimental Results

For the experiments, we have considered the sequence Foreman (QCIF format and 15 Hz). The group of pictures (GOP) size is set to two (even frames are encoded as key frames and odd frames as Wyner-Ziv frames). The number of diagonals in the index assignment matrix, which controls the amount of redundancy of the MDSQ, is set to 3.

As one can see, when using almost the same bitrate as the single description coding scheme, the proposed balanced MDC scheme can achieve a good quality when the two descriptions are received or when only one description is received.

PSNR = 37.1 dB, bitrate = 420.4 kbs Fig. 2 - Sequence obtained with the single description coding scheme. Download the AVI file	PSNR = 34.17 dB, bitrate = 412.6 kbs Fig. 3 - Sequence obtained with the proposed MDC scheme at the central decoder. Download the AVI file
PSNR = 32 dB, bitrate = 225.4 kbs Fig. 4 - Sequence obtained with the proposed MDC scheme at the side decoder 1. Download the AVI file	PSNR = 32.1 dB, bitrate = 191 kbs Fig. 5 - Sequence obtained with the proposed MDC scheme at the side decoder 2. Download the AVI file

References

• Olivier Crave, Christine Guillemot and Béatrice Pesquet-Popescu, Multiple description source coding with side information, EUSIPCO 2008, 16th European Signal Processing Conference, Lausanne, Switzerland, August, 2008.