Efficient low bit rate algorithm for image coding based on VQ and entropy coding

Vector quantization (VQ) is an efficient technique for speech and image coding because of its superior rate-distortion performance. The design of an optimal codebook has been an area of keen research interest in recent years. In this paper, a novel equidistortion principle based clustering algorithm is proposed which is found to yield better codebooks than existing ones. Further, a new mechanism for entropy coding of the codeword addresses is proposed, which decreases the output bit rate by a factor of almost two compared to ordinary VQ. This is achieved by an address mapping technique, which exploits the intervector correlation to translate the original set of codeword addresses to a set of reduced entropy addresses. A two-stage image coding scheme is proposed based on the above ideologies. Simulation results on still monochrome pictures of the USC database show that the proposed scheme gives good coding performance at low bit rates.