Influence Domain Constraints In Adaptive Geodesic Active Contour Model

Image segmentation, as a key technology of image process/analysis, is defined as the process to separate one image into several regions based on regional similarity and inter-regional differences. In recent years, Active Contour Model has gained significant progress. The major contributions in this paper are as follow:1. A new mathematical model called Geodesic Active Contour with Adaptive Neighborhood Inference (GAC-ANI) is presented. This new method defines a narrow band near the isocontour, in which the front points are replaced by the integration of all points within its inference domain. Thus it is robust to noise and capable of handling boundary leakage problem. The GAC-ANI is successfully used in IR image segmentation in the practical experiments.2. To eliminate the sensitivity of the initial contour and expand the capture range of boundaries, the gradient vector flow (GVF) and adaptive balloon force are introduced to traditional GAC model as the external energy. Besides, the stop problem of the front curve that the tangent vector parallels to the GVF is also resolved by applying the adaptive balloon force.3. As for the numerical method, the meshfree method is introduced to solve the Level Set equation. To evaluate the accuracy property of our method, a comparison between meshfree and the finite difference is also performed in this paper.