Quantisation

Real numbers, and consequently DCT coefficients, are very expensive in terms of transmission efficiency. Therefore, quantisation is applied to reduce the representation accuracy of the transform, and delete, or set to zero, the frequently appearing near-zero coefficients. This is the point where the information leakage that makes the scheme lossy resides, as, in the inverse process at the decoder, it is not possible to restore precisely the DCT coefficients, but only an approximation of them, depending on the selected quantisation parameters.

DCT coefficients quantisation does not have the effect of, for example, DPCM samples quantisation, where a particular quantised sample affects relatively few pels in the reconstructed image, as in the former all pels of the block are affected. This leads to a spreading of the quantisation error, and much better subjective quality of the reconstructed images [NH95]. However, the unavoidable loss of information produces distortions at the block boundaries, or `block artifacts,' which are emphasised by the limited precision of block-based motion estimation, usually used in conjunction with DCT.

Hosted by www.Geocities.ws