Image Segmentation Using an Active Contour Model Based on the Difference Between Local Intensity Averages and Actual Image Intensities

Abstract

The local intensity fitting active contour models can handle inhomogeneous images, but they suffer from the shortcomings of poor performance in segmenting images with severe intensity inhomogeneity and being sensitive to initializations. To overcome these problems, we put forward a robust active contour model by introducing two adjustment coefficient functions. The energy functional of the proposed model is presented by integrating the local fitting term and two adjustment coefficient functions. The local fitting term is defined by introducing two local fitting functions that approximate the image intensities inside and outside of the contour. These two adjustment coefficient functions, which improve the segmentation performance and enhance the robustness to initialization, are constructed by utilizing the Sigmoid function as well as the difference between local intensity averages and image actual intensities. The results of the experiments on synthetic and real images demonstrate that the presented model not only is capable of handling intensity inhomogeneity better under more flexible initializations but also takes less time in comparison with other region-based models. Furthermore, these two adjustment coefficients can be employed to other local intensity fitting models to enhance the robustness to initialization and to decrease the segmentation time.