A Variational Level Set Method for Multiple Object Detection

摘要

A novel variational level set method for multiple object detection is presented, which uses n - 1 level set functions for n - 1 objects and the background without overlapping and vacuum problems. The energy functional includes three parts. The first part is a parametric region-based model via generic image noise distributions, the second part is the classic edge-based model, the third part is a term used to enforce the constraints of level set functions as signed distance functions. Characteristic functions for region partitioning are written in a unified form using Heaviside functions of level set functions. Some intermediate terms in evolution equations are extracted in a unified form for simplification of expressions and computation efficiency. The corresponding semi-implicit schemes are derived and used to some examples for segmentation of synthetic and real images to validate the method suggested in this paper.