Study On The Radial Basis Function Method For Image Segmentation

Image segmentation is important in image processing. The level set method (LSM) has been widely applied in image segmentation. The principal idea is to implicitly express the interface contour as the zero level of a level set function, and then deduce a nonlinear evolution equation. In the past, special time-consuming finite difference methods are generally used to solve the evolution equation numerically. In traditional LSM, segmentation results usually rely on manually initial contours. Besides, re-initialization procedures are used periodically to restore the regularity and stability of the level set function. Re-initialization procedures need complicated partial differential equation solving schemes and may incorrectly move the zero level set away from its original position.This dissertation focuses on the LSM-based image segmentation. A radial basis function method is developed for image segmentation. Some drawbacks of traditional LSMs are addressed. The main contents are as follows.Firstly, by using radial basis functions to interpolate the level set function, the nonlinear evolution equation reduces to a system of ordinary differential equations (ODEs).Secondly, Euler’s scheme is applied to tackle these coupled ODEs, and iteratively computational formulas are given in detail.Finally, the capability of the current method is demonstrated through segmentation of some synthetic and real images.Compared with traditional LSMs, the current image segmentation method can free of initialization, and avoids re-initialization procedures. Experimental results indicate that this method can segment images efficiently even without any initial contour.