Adaptive nonlinear multivariate image filtering for mixed noise removal

An adaptive nonlinear multivariate image filtering method with particular applications to color image processing is presented. A multivariate estimate is formed according to the rule that it is the input sample in the current window which has the minimum weighted sum of squared distances to three multivariates, namely, the mean, the marginal median and the center of the samples within the current window. Adaptation is achieved by varying the weights in the weighted sum. A simplified version where the output is just a weighted sum of the above three multivariant is introduced. Both filtering schemes are compared with other nonlinear multivariate filtering schemes. The results show that the methods provide rather good noise attenuation and detail preservation