CONVEX PROJECTIONS BASED EDGE RECOVERY IN LOW BIT-RATE VQ

This paper proposes an implementation for vector quantizers (VQ) with very small codebooks (i.e., 30-32 codevectors) for compressing graysca le images. The technique uses a convex projections (CP) based algorith m for iterative restoration of edges, as part of the decoding process. The objective of this approach is to code the edge blocks vestigially by drastically reducing the number of edge vectors in a codebook. Thi s will result in a large reduction in codebook size and hence fast sea rches. Also this method works better on images outside the training se t since encoding is less dependent on the edges.