A fast narrow band level set formulation for shape extraction

Shape modeling is an active area of research in Computer Graphics and Computer Vision. Shape models aid in the representation and recognition of arbitrarily complex shapes. This paper proposes a fast and computationally efficient narrow band level set algorithm for recovering arbitrary shapes of objects from various types of image data. The overall computational cost is reduced by using a five grid point wide narrow band applied on a variational level set formulation that can be easily implemented by simple finite difference scheme. The proposed method is more efficient and has many advantages when compared to traditional level set formulations. The periodical reinitialization of the level set function to a signed distance function is completely avoided. Implementation by simple finite difference scheme reduces computational complexity and ensures faster curve evolution. The level set function is initialized to an arbitrary region in the image domain. The region based initialization is computationally more efficient and flexible. This formulation can form the basis of a shape modeling scheme for implementing solid modeling techniques on free form shapes set in a level set framework. The proposed method has been applied to extract shapes from both synthetic and real images including some low contrast medical images, with promising results.