Noise reduction method for image processing using genetic algorithm

Growing computer networks and multimedia systems require more efficient image processing methods. Therefore, many image processing, methods have been proposed to reduce the load of systems. Especially, image noise reduction methods are important to preserve transparency of network communications as a preprocessing filter of the other methods, e.g. image recognition, compression. Some of these noise reduction methods are based on minimizing the dependence among input signals to separate a noise component. A noise component is usually independent of the other signals. Under such circumstances, we have developed a new method to separate a noise component which directly minimizes the Kullback-Leibler divergence by a genetic algorithm (GA). The Kullback-Leibler divergence measures statistical dependence among input signals and is expressed in the product of the probability density function and the marginal probability density functions of input signals. This measure is lower when input signals have lower dependence from each other. Therefore, finding the transformation of input signals which minimizes this measure is equivalent to separate independent noise components from the noise mixed input signals. We have adopted a genetic algorithm to minimize the Kullback-Leibler divergence