LBG algorithm is classical algorithm in Vector Quantization

which was proposed in 1980 by Y. Linde

A. Buzo

and R. M. Gray. Used this algorithm

an acceptable performance codebook can be get in the acceptable time

it has better performance than Scale Quantization that was proved by Shannon. But few outlines vector of the training image effect the distribution of the codeword in the codebook. To few vectors

it maybe noises

but have large codeword number in the codebook. And decrease the body vectors' codeword in the codebook. That decreases the compress ratio and makes the reconstruct image worse

the advantageous of the Vector Quantization can't be explained adequately. Different people used this algorithm with different image can get different compression ratio. Design the Vector Quantizer based on robust statistic can improve it. Decrease the outlines vectors

improve the center vector effect in the codebook

it can decrease the relativity of the codebook

made the distribution of the codeword of the codebook more economical and bring on the compression ratio. Theoretical analysis and simulation experimental results presented in the paper show that this method can obtain good reconstruction image quality and high compress ratio. It is improved in both subjective and objective.1