Encoding of still pictures by the five-dimensional vector quantization and zero-tree implementation

This paper presents research results on image compression by wavelet transforms and 3-level quad tree coding. We decompose an input image to (3×3+1) subband images using 2-D discrete wavelet transforms three times, then arrange the wavelet coefficients to 3-level quad trees. Next we divide the quad trees into zero trees and non-zero trees. After that we draw a root map of zero trees and encode the map by a special MH (modified Huffman) code. The process described above enables us to encode 0-amplitude wavelet coefficients very efficiently. For encoding non-zero trees, we propose to separate a 3-level non-zero quad tree to five 2-level quad trees, then encode the 2-level quad trees by 5-dimensional vector quantization. Detailed simulation results on the encoding of the root map and investigated results on statistical properties of non-zero trees are presented