A novel active contour model driven by statistical and gradient information is proposed in this paper. The model not only efficiently utilizes the gradient information of an object

which is in favor of fast and accurate location of boundaries

but also makes full use of the statistical information

including the global and local region information

which makes our method robust to noise. The use of the local region information makes the method free from intensity inhomogeneity of images

and the use of the global information helps to avoid the evolved contour trapping into local minima. Therefore

the initial contour can be set anywhere. Finally

the level set function is regularized by a Gaussian convolution kernel

which avoids an expensive computational re-initialization or regularization of the conventional models. Experimental results show that the proposed method can accurately and efficiently segment the homogenous images

as well as the inhomogenous images

with the initial contour set anywhere. Furthermore

the model is robust to noise.