PRISM: A Video Coding Architecture Based on Distributed Compression Principles

We introduce PRISM (Power-efficient, Robust, hIgh-compression, Syndrome-based Multimedia coding), a video coding paradigm underpinned by the principles of distributed source coding from multi-user information theory. PRISM, which rests heavily on channel coding concepts, represents a radical departure from current state-of-the-art video coding architectures, including those driving popular standards like MPEG and H.263, that are based on a motion-compensated prediction framework. These are hampered by: (i) a rigid computational complexity partition between encoder (heavy) and decoder (light); (ii) high fragility to drift between encoder and decoder in the face of prediction mismatch, e.g. due to channel loss; and (iii) being part of a standards environment requiring a bulky encoding format that makes innovating within the standard cumbersome. In stark contrast, PRISM’s architectural goals are: (i) to have a flexible distribution of computational complexity between encoder and decoder without compromising compression efficiency; (ii) to have inbuilt robustness to drift between encoder and decoder due to channel loss; and (iii) to have a light yet rich encoding syntax that can be standardized in a way that permits for creative growth and unencumbered innovating within the standard. In this paper, we describe a specific implementation of PRISM corresponding to a near reversal in codec complexities, with a rough swap in encoder and decoder complexities vis-a-vis today’s codecs. This leads to a novel light encoder and heavy decoder paradigm. Specifically, this involves moving the expensive motion search component of the video codec from the encoder to the decoder. Simultaneously, PRISM offers a low-latency, easily-tunable natural immunity to the drift problem. In a nutshell, the ideal is to marry the encoding lightness and robustness of intraframe coding (so-called motion JPEG) with the compression-efficiency of interframe coding (full-motion MPEG). Our preliminary simulations for a specific block-DCT based implementation of PRISM confirm the promise of realizing this ideal. Despite being based on very simple codes, our preliminary experiments reveal the strikingly superior performance potential of PRISM to state-of-the-art prediction-based codecs under the regime of packet/frame drops, as well as how surprisingly closely, considering this is a new untested paradigm, it approaches their subjective performance even in the zero-channel-loss regime.