Vector Quantization in MPEG

摘要

The Motion Picture Experts Group (MPEG) video coding algorithm is a promising coding technique for coding full-motion video.The MPEG standard exploits multiple compression techniques with most of the compression implemented by using motion compensation,DCT transformation and quantization,and entropy coding.The DCT transform converts an 8×8 block of pixel values to an 8×8 of horizontal and vertical spatial frequency coefficients and assigns 8×8 frequency coefficients to a linear vector from low to high frequencies by a zigzag scan.Finally,the quantized linear vector is entropy coded.In this paper,the linear vector is divided into several subvectors which were coded by the vector quantization technique.Since the low DCT frequencies are more visually important than the high frequencies in an image,more bits are assigned to low frequencies than to the high fre quencies.We give a bit rate allocation algorithm which determines how to allocate a given quota of bits among the subvectors.Experimental results show that the vector quantization (VQ) encoder produces higher quality video signals than the MPEG eneoder with a given quota of bits in terms of the mean value of SNR.However,it is difficult for the VQ encoder to compress video signals with fixed quality.