New methods for signal and image processing by information theoretic divergence

点击翻译按钮获取中文摘要

1. We proposed a unified class of equivalent bandwidths of a random process by using Renyi's a-order entropy. It is useful to evaluate randomness of a random process.2. We investigated global thresholding of a image. In addition to traditional one-step methods, we proposed a new idea of the two-steps method. In the first step of the method, you estimate the optimum model probability distribution that the grey-level variable obeys and in the second step you decide the optimum threshold to binarize the image. We propose a new binarization-oriented method for model distriobution fitting where we introduce class labels as hiden-variable and, instead of the traditional divergence measure, information theoretic mutual information measure is used. Simulation experiments shows our methodes are very powerful.