Convex Background Removed Model for Image Segmentation Using the Split Bregman Method

摘要

A novel Convex Background Removed (CBR) model is proposed in this paper. The model aims to detect objects by removing the background. With application of an approximation of the 1-D Heaviside function which takes 0 for negative and 1 otherwise, the convex energy of the CBR model is obtained. Just for this convexity, the model can avoid the common drawback that most models get local minima, which makes it insensitive to the initialization. At the computational level, we employ a fast minimization technique, the split Bregman method, to minimize the convex energy function. The high accuracy and efficiency of the proposed model are demonstrated by the experiments. Moreover, it shows that the CBR model is robust to noise.