In this chapter, we investigate two popular techniques for error-resilient H.264/SVC video transmission over packet erasure networks, i.e., layered video coding (LVC) and scalable multiple description coding (SMDC). We compare the respective advantages and disadvantages of these two coding techniques. A comprehensive literature review on latest advancement on SMDC is provided. Furthermore, we report new simulation results for the novel two-dimensional scalable multiple description coding (2-D SMDC) scheme proposed in our previous work (Xiang et al., 2009). The 2-D SMDC scheme allocates multiple description sub-bitstreams of a two-dimensionally scalable bitstream to two network paths with unequal loss rates. We formulate the two-dimensional scalable rate-distortion problem and derive the expected distortion for the proposed scheme. To minimize the end-to-end distortion given the total rate budget and packet loss probabilities, we need to optimally allocate source and channel rates for each hierarchical sub-layer of the scalable bitstream. We consider the use of the Genetic Algorithm to solve the rate-distortion optimization problem. The simulation results verify that the proposed method is able to achieve significant performance gain as opposed to the conventional equal rate allocation method.
TopIntroduction
In recent years, the increasing demand for mobile broadband access to multimedia applications and services has motivated enormous interest in media streaming over Internet. Media streaming is characterized by high data transmission rate and low delay constraint. Therefore, its applications are highly susceptible to packet delay and delay jitter. However, packet erasure networks such as Internet only offers best-effort service, and thus cannot guarantee bandwidth and delay. Conventional approaches for tackling packet loss such as retransmission is not effective for streaming media due to real time nature of the service. Thus, additional mechanisms are needed to provide streaming media delivery over packet erasure networks.
Traditionally, layered video coding (LVC), which is also known as scalable video coding (SVC) with transport prioritization (Chakareski, 2005), is the most popular and effective scheme for video transmission over error-prone networks. In LVC, a raw video sequence is coded into a base layer that provides a coarse level of visual quality and can be decoded independently, and multiple enhancement layers that refine the base-layer visual quality and are nevertheless useless alone. SVC has the advantage of enabling the media providers to generate a single embedded bitstream from which appropriate subsets can be extracted to meet various requirements of a broad range of clients. SVC is thus essential for multicast applications, where a variety of end users with different capabilities such as bandwidth and processing power receive different presentations of the same media content. In general, temporal scalability, spatial scalability, and signal-to-noise-ratio (SNR) scalability are among the most common scalability mechanisms for SVC. These mechanisms provide for frame rate, spatial resolution, and video quality adaptation to adjust the video source in accordance to device capability and channel bandwidth.
However, the fact that enhancement layers provide little refinement information if the base layer is not received or decoded correctly, is the root cause that scalable video bistreams are extremely vulnerable to transmission errors. Therefore, in an error-prone environment such as the Internet, providing error resilience against packet errors is as important as enabling scalability. LVC is usually used in conjunction with unequal loss protection (ULP) (Mohr et al., 2000), which provides the base layer with the highest level of channel error protection through the use of forward error correction (FEC) coding.
The most weakness of LVC lies in its excessive reliance on the correct receipt and decoding of the base layer. As an alternative, multiple description coding (MDC) (Gamal & Cover, 1982), (Goyal, 2001) has recently emerged as a promising alternative technique for providing graceful degradation of performance in the presence of channel noise. The essential idea underlying MDC is to generate multiple (> 2) independent descriptions of a source such that each description independently describes the source with a certain desired fidelity. When more than one descriptions are received, they can be synergistically combined to enhance the quality. Therefore, the distortion in the reconstructed signal decreases upon the receipt of any additional descriptions, and is lower bounded by the distortion attained by single description coding.
In recent years, a plethora of different MDC methods have been proposed in the literature. The pioneering MDC scheme termed multiple description scalar quantization (MDSQ) (Vaishampayan, 1993) was proposed by Vaishampayan in 1993. The scheme minimizes the central distortion through jointly designing the quantization levels and the index assignment, subject to prescribed constraints on the side distortion and the rate. This principle has been generalized to multiple description trellis and lattice vector quantization in (Hafarkhani, 1999), (Vaishampayan, 2001), respectively. Multiple description transform coding (MDTC) is another popular class of MDC paradigms that employ pairwise correlating transforms (Goyal and Kovacevic, 2001).