Master Title: Wavelet Network For Image Compression

Abstact:

                For many years, we have witnessed an increasing growth in the need of numeric pictures (whether stationary or animate) in numerous fields such as telecommunications, multimedia diffusion, medical diagnosis, telesurveillance, meteorology, robotics, etc. However, this type of data represents a huge mass of informations that are difficult to transmit and to stock with the current means. Thus, it was necessary to have new techniques known as picture compression.

 Different methods, whether without loss of information as Shannon-Fano, Huffman, … or with loss of information not perceived by naked eye as the transformed discreet cosine (TDC), the standard Jpeg, the  wavelets, … have been presented.

 In this paper, we have brought our contribution to the utilization of the wavelet analysis in the artificial neural networks i.e wavelet networks. We have described the architecture of this network and its training algorithm, and then we have presented the stages and the principle of picture compression using such a network. To perceive the performances of such a system, a graphic prototype using Matlab has been developed to validate the developed algorithms and to test the different follow-up approaches. Finally, to appreciate numerically the results and to check the efficiency of the proposed process, we have determined some measures of fidelity and cleared tables and assessment curves depending on several criteria on which our wavelets network depends like: the type of the wavelet’s family, the number of these wavelets, the number of iterations, etc.

Key-words : Picture compression, Wavelets, Neural networks, Wavelet networks.