Image sequence coding by three-dimensional block vector quantisation

In the paper results on the use of vector quantisation for interframe image coding for compression are presented. The initial stage in vector quantisation is the formation/extraction of vectors from the data source. Two vector formation schemes, generalized from intraframe image coding, are proposed and evaluated on a test image sequence. The first, direct three-dimensional block vector quantisation, applies vector quantisation to small 3-dimensional blocks of pixels extracted from sequences of image. In the second, the 3-dimensional Hadamard transform is first calculated on a block basis and is followed by vector quantisation of a subset of the Hadamard coefficients. Significantly improved results are obtained in the second case, as the components in the vectors are decorrelated in both the spatial and temporal directions. Three training sets are used for the simulations, and results suggest that adaptively updating the codebook could result in improved performance.