Fuzzy automatic classification and without a prior knowledge - mean shift application to MRI brain images segmentation [articol]

Abstract

The objective of this paper is to propose a blind segmentation method able to localize all relevant objects in medical images by using a fuzzy classification based mean shift algorithm. To achieve this, we have to build a cartography of attributes consequent upon images characterization. The objects localization is realized by searching modes from a point sample distribution through mean shift procedure. In order to obtain an image automatic segmentation, the approach is joined at a fuzzy classification based fuzzy c-means (FCM) approach. The fuzzy insertion here allows to take into account imprecision related to information extraction which is necessary for a region classification. Actually, the obtained results by our approach are very encouraging and show an accurate segmentation compared to others supervised techniques.