Block-Adaptive Lattice Vector Quantization in Image Coding

Abstract

A subband image coding algorithm is described based on lattice-based spherical VQ and lattice-based pyramid VQ. The algorithm partitions a subband (or wavelet) decomposed image into blocks of various sizes, depending on their energy and complexity constraints on the enumeration encoding of lattice codevectors. Each block is lattice vector quantized and encoded using a product code. The algorithm is simple and effective, exploiting energy clustering in a wavelet transformed image. Using the integer lattice, the algorithm provides performance slightly better than the Set-Partitioning Embedded Block (SPECK) algorithm, and is competitive with JPEG2000 and the set partitioning in hierarchical trees (SPIHT) algorithm.