A Kernel-Based Post-Process for Image Segmentation Using GVF Snake

This paper presents a kernel-based post-process for the segmentation of multiple objects using gradient vector flow (GVF) snake (active contour) algorithm. GVF snake has stronger convergence proficiency to boundary convexities and concavities than traditional snake. However, because of the nodes affiliation to each other and mislead of external and internal forces thatare used in GVF snake, some of the nodes remain between trueboundaries as they push toward true boundaries, and resultsin an incorrect segmentation. Our algorithm has two steps. First we decomposedeach pixel of image toits lower size and composed a new image with new grid sizes, then we used an adaptive kernel-based standard deviation calculation for each node of snakes to evaluate its accuracy if it is true segmented result or not. We have tested the proposed method on some sorts of synthetic images and a gray-level real satellite image thatwas captured with Orbview3 panchromatic band with 1m resolution and we have achieved considerable results. 