Bias field estimation and segmentation of MR image using modified fuzzy-C means algorithms

The brain Magnetic Resonance (MR) image has an embedded bias field. This field need to be corrected to obtain the actual MR image for classification. In this paper, we have proposed three new schemes to simultaneously estimate the bias field and obtain segmentation. These algorithms are modification of Ahmed et al.'s [4] Bias Corrected FCM (BCFCM) algorithm. The first proposed scheme considers the weighted typicality measure for the data set. This results in Bias Corrected Possibilistic FCM (BCPFCM) algorithm. Besides, to improve the segmentation accuracy, we have considered the joint effect of weighted membership and typicality of the neighborhood pixels resulting in Bias Corrected Possibilistic Neighborhood FCM (BCPNFCM) algorithm. The third notion considers the weighted membership and weighted typicality separately weighted for the neighboring pixels in addition to the pixel in consideration. The corresponding algorithm is Bias Corrected Separately weighted Possibilistic Neighborhood FCM (BCSPNFCM) algorithm. The proposed algorithms have successfully been tested on synthetic image and also found to produce appreciable results in case of axial brain MR image data as compared to BCFCM algorithm.