A neural and morphological method for wavelet-based image compression

Image compression using the wavelet transform has several advantages over other transform methods. However, wavelet-based compression methods require not only the encoding of the significant coefficients, but also of their positions within the image. The paper presents a wavelet-based image compression method where the significance map is pre-processed using mathematical morphology techniques to create clusters of significant coefficients. It is then encoded using a competitive neural network whose training rule was developed to take advantage of some properties of this kind of problem. Some experimental results are presented to validate the competitive learning rule and other components of the method