A modified Kernelized Fuzzy C-Means algorithm for noisy images segmentation: Application to MRI images

Image segmentation is a low-level processing operation; it is the basis for many applications such as industrial vision and medical imaging. Segmentation provides a partition of the image by gathering pixels with similar grey levels in the same class. The main problem of this algorithm is that it does not take into account the image topology; it is based only on the pixels values. Thus, it is very sensitive to noise and inhomogeneities in the image moreover, it remains dependent on the initialization of the cluster centers. In general the clustering algorithm chooses the initial centers in a random manner but the cluster centers initialization, using "Expectation Maximization" algorithm allows an optimal choice of these centers. To account for the topology of the image, the statistical parameters of a window around the pixel are considered. These a priori are used in the optimization of the cost function. The application to MRI images from the Brain Web public database, with different noise levels, shows the performance of the proposed approach.