Local Intensity Fitting Active Contour Model Based on Gradient Constraint

To deal with the drawback of being sensitive to the initial contour in local region based active contour models, a novel local intensity information and gradient information based active contour model is presented. The energy function for the proposed model consists of three terms: local regional information term, gradient information term and length term. Besides, an energy minimization model with inequality constraints is proposed using those terms. Firstly, local intensity energy and contour length energy are linear combined as the objective function, which can drive the level set function towards the boundaries of the objects. Secondly, an inequality constraint is constructed by gradient energy. Then the optimization problem with inequality constraints is transformed to an unconstrained optimization problem by penalty function method. This term can make the level set function locate in the neighborhood of the boundaries, which can overcome the disadvantages of the oriental model. Finally, Gaussian convolution is introduced to regularize level set function as a signed distance function, which makes the evolution more stable and avoids reinitialization at the same time. Experiments conducted on some synthetic and real images verify that the proposed model is robust to the selection of the active contour model and can segment images with intensity inhotnogeneity.