New classified vector quantization with quadtree segmentation for image coding

摘要

A new variable block-size, classified vector quantization (VQ) scheme for image compression is presented. A quadtree segmentation is employed to generate blocks of variable size according to their activity. Inactive areas are coded by a predictive VQ of low encoding rate, while active areas are coded by an edge-classified VQ to avoid edge degradation. Both the segmentation and the edge classification are determined by a simple universal contrast identifier. It is shown that the performance of this scheme is competitive with other classified VQ-based approaches, while the coding complexity is relatively low.