Dynamic address-vector quantisation of RGB colour images
Yili Feng, Nasser M. Nasrabadi · IEE Proceedings I Communications Speech and Vision · 1991
Address-vector quantisation (A-VQ) is a new class of vector quantisation (VQ) technique that exploits the statistical redundancy between the neighbouring blocks, to reduce the bit rate to a much smaller value than a standard vector quantiser can achieve. The underlying idea of A-VQ is that the interblock correlation is exploited by encoding a group of neighbouring VQ addresses together. The VQ addresses are obtained by vector quantisation of each individual block. An address codebook, which consists of a set of address codevectors, is then used to encode the neighbouring VQ addresses together; this process introduces no extra coding error since it is a vector dimensionality reduction process. In the paper, the A-VQ coding technique is extended to encode RGB colour images by utilising the interblock as well as the intercolour correlation. Also, a modified A-VQ technique named dynamic address-vector quantisation (dynamic A-VQ) is introduced for the encoding of colour images. In the dynamic A-VQ, the encoder dynamically tracks the local characteristics of the input image. Experimental results show reconstructed images at bit rates of 0.5–0.6 bits per pixel. The resulting bit rate is less than a standard VQ coding technique by a factor of at least two, with the same quality of reconstructed images.