Abstract
Normalized AC transform coefficients of picture signals are well modeled as having a spherically symmetric distribution. For the distribution, quantizers whose code-words are arranged on the surface of concentric hypersphere will show excellent performance with its low encoding complexity. Permutation codes, developed by Berger and others, having this property, require no multiplication for encoding, and can directly quantize a very high dimensional input. Permutation codes, however, cannot show a satisfactory performance for high rates or low dimension, and require unfeasible operation precision for encoding a codeword's index. To cope with these problems, we developed new improved permutation codes, formed an algorithm for encoding a codeword's index, which is performed by only integer operations with limited operatdon precision, and incorporated improved permutation codes into discrete cosine transform image coding. The simulation results demonstrate that improved permutation codes can efficiently quantize transform coefficients of a high sequency.
