Linear predictive coding of imagery for data compression applications

Extremely efficient compression algorithms can be devised for digital imagery data via an extension of the linear predictive coding methods utilized extensively in speech processing. When appropriately employed, these methods introduce minimal distortion. Such realizations can operate in real time, completely in the spatial domain, and are capable of reducing imagery storage requirements by an order of magnitude. This paper describes such an extension of linear predictive coding techniques for imagery data compression applications both theoretically and experimentally. This realization includes several heretofore uncombined and novel approaches to the imagery compression problem including: 2-D lattice filter prediction, adaptive quantization, and entropy coding. System implementation shows proof of feasibility and favorable performance in comparison with alternative transform techniques using the standard Minimum Mean Square Error [MMSE) fidelity criterion.