Abstract: To improve the performance of the active contour segmentation algorithm on natural images, a novel segmentation algorithm is proposed. First, combining the level set with the total variation, an edge-preserving smoothing segmentation model is constructed. Then a kind of clustering algorithm is employed to learn the balance parameter adaptively to avoid the level set curve converges at the local optimal point. At last, according to the different smoothing components with different segmentation regions, the segmentation smoothing convergence function based on regional confidence is designed to solve segmentation curve vanishes. Experimental results show that the score of this algorithm is higher than that of the traditional active-contour-based segmentation algorithmsfor the real images, and the algorithm is insensitive to texture and noise.