S. Hatami - Department of Electrical and Computer Engineering University of Tehran
M.J Yazdanpanah
B. Frozandeh
O. Fatemi

The increased demands for image storage in computer systems and transmission in communication systems have magnified the importance of the demand for signal and image compression algorithms respectively. We have focused on Vector Quantization (VQ), as a well-known compression technique, which has been widely used in many speech and image coding systems. Algorithms such as LBG and SOM (a neural network (NN) algorithm) are used towards to find a proper codebook for a given training data in VQ. We have also computed a modified version SOM called SFS_HSOM. In this paper, we used four techniques to improve the reconstructed image quality up to 130% and to decrease training and encoding time.

Vector Quantization, Neural Networks, Self_Organizing Feature Map