A Novel Optimizer Based on Particle Swarm Optimizer and LBG for Vector Quantization In Image Coding

This paper presents an optimizer based on particle swarm optimization and LBG (PSO-LBG) for vector quantization in image coding. Three swarms, including two initial swarms and one elitist swarm whose particles are selected from two initial swarms respectively, are applied to find the global optimum. At each iteration of a swarm´s updating process, particles perform the basic operations of PSO, but with smaller parameter values and population size compared with conventional PSO, followed by the well-known vector quantizer, i.e. LBG algorithm. Experimental results have demonstrated that the quality of codebook design using this optimizer is much better than that of fuzzy k-means (FKM), fuzzy reinforcement learning vector quantization (FRLVQ) and FRLVQ as the pre-process of fuzzy vector quantization (FRLVQ-FVQ) consistently with shorter computation time and faster convergence rate. The final codevectors are scattered reasonably and the dependence of the final optimum codebook on the selection of the initial codebook is reduced effectively.