Unsupervised optimization of morphological filters for noise removal in texture images

抄録

A method of optimizing structuring elements of morphological filtering for noise removal in texture images without any example of noiseless images is proposed. This method is based on the property that texture images are composed of repetitive appearance of microstructures, and the method applies the estimated shape of the microstructure as the optimal structuring element of the morphological opening and closing. The robustness of the estimation under the existence of noises is improved by introducing limitations to the pixel value distribution on the structuring element in the process of optimization, and iterating the optimization with modifying the size of the structuring element.