SEGMENTATION OF MEDICAL IMAGES USING LEVEL SET METHOD

Abstract

This thesis proposes a novel theoretical framework, namely the curve evolution implementation of the Mumford—Shah objective function for dealing with the image segmentation. This frame work is based on the mathematical properties of the operator. The active contour model without edges based on the Mumford-Shah function is fully dependent on the initial position of the curve; it crosses the required boundaries if the number of iterations are more, and also results in unnecessarily splitting of the curve In another method, the initial curve which is necessary for the level set method is created by the randomized Hough transform (RHT) which overcomes the limitations of the Hough transform and some morphological operators are also used in pre-processing stage along with the randomized Hough transform. The initial curve which is created by this transform is the basis for total level set method and it effectively overcomes the limitations of the basic method. The RHT based method does not require reinitialization of the distance function in every step and so speed of this method is increased. With creation of the ellipse in proper position by randomized Hough transform, the required fetal head is detected within few iterations. The performance of this method is measured with the error metric called Hausdorff distance. The speed function of this model is based on the image region and "boundary gradient information. The method proposed here combines the randomized Hough transform and active contours without reinitialization to overcome the limitations faced by traditional methods.