An image coding scheme using the discrete cosine transform is analyzed when the transform coefficients are vector quantized. The coding method is based on the known scheme proposed by Chen and Smith (1977) which sorts the picture blocks into classes according to the level of image activity. The coding scheme is modified to allow for vector quantization of the ac coefficients, in particular a pyramid vector quantizer (PVQ) is used. This is based on the statistical and geometric properties of a Laplacian source which, in fact, is the best model for the ac coefficients of the two-dimensional discrete cosine transform (2D-DCT) of an image. A method for forming almost statistically independent vectors is also suggested and improves quantization performance. Images are encoded with both the PVQ and standard scalar quantizer transform coders, demonstrating that the PVQ coder reduces the mean square encoding error and improves image quality. In particular, emphasis is given to how the use of fractional bit rates affects the objective and subjective gains obtained. The results presented (i.e. mean square error values and printed images) have been obtained experimentally, working with a statistical criterion in a group of images whose size was in accordance with the 50 Hz CCIR Recommendation 601 Standard.>

A new effective adaptive coding scheme for data compression of images is presented. It is based on a reduction of the spatial, temporal and statistical redundancy. Inter-frame motion compensation technique was used to take advantage of the image temporal correlation. A unitary 2D-DCT was performed on each (8x 8) pixel block in order to exploit spatial correlation and in order to decorrelate the block elements and to make the relative quantization error subjectively less disturbing. The Rate Distortion Function of a Laplacian source was used as a classification strategy by relating a block to a coding bit rate in accordance with its energy. The a.c. transform coefficients were vector quantized by using the pyramidal vector quantizer based on the cubic lattice. Compression ratios in the 8 — 12 range give a distribution quality for colour image sequences according to the CCIR Rec. 601 (12-20 Mbit/sec) or HDTV sequences (50-70 Mbit/sec). Experimental results obtained by computer simulations are presented and the influence of the motion compensation and the vector quantization on the objective error are analysed.