An improved LBG algorithm for image vector quantization

An improved LBG algorithm for vector quantization is introduced in this paper. Its basic idea is classifying the input vectors based on space partition with a variational distance threshold. Firstly, set the value of distance adjusting factor and cipher out the initial distance threshold, and then the input vectors will be classified into the corresponding cells or to be new clustering vectors. After that, select the clustering vectors which have less input vectors in their cells than average vector number, and delete these cells. Thirdly, update the input vector set and decrease the distance threshold, continue to the next iteration until the number of selected clustering vectors meet requirement. Finally, consider these clustering vectors as the initial codebook for LBG algorithm. The simulation results show that the reduction of iteration times is outstanding and the effect of reconstructed images is heightened obviously.