Directional Preserving Gradient Vector Field for Active Contour Models

This paper presents a novel dynamic external force for snakes named directional preserving gradient vector field (DPGVF). Firstly, it proposes the concept of boundary direction preserved using the features of the Jacobian, which makes the segmentation results better. Secondly, it makes use of the gradients of gray level images and defines positive and negative boundaries in horizontal and vertical directions respectively.The experimental results show that the method largely prevents interference from the initial contour to the boundary closely and weak boundary, leakage of the boundary clearance and other issues. Additionally, the DPGVF Snake has larger capture ranges and better initialization. It has excellent segmentation results for complex medical images.