Generalized gain-shape vector quantization for multispectral image coding

This paper proposes a new encoding scheme that generalizes the Gain-Shape Vector Quantization technique and takes advantage of the distinctive features of multispectral images to encode them at very low bit rate, with a satisfactory reproduction quality and low complexity. Each codevector is obtained as the Kronecker product of a gain codevector and a shape codevector, which reduces both the memory requirements and code-book design complexity. Besides, the encoding complexity is also greatly reduced by resorting to a fast encoding algorithm.