Local Dispersion Active Contour Model for Image Segmentation with Intensity Inhomogeneity

Image segmentation is an important procedure in image processing and computer vision, active contour model methods have been widely used in image segmentation. Intensity inhomogeneities often occur in real-world images, and it may lead to serious misclassifications by intensity-based segmentation algorithms that assume a uniform intensity. In order to overcome the difficulties, a local dispersion-based active contour model for image segmentation is proposed. Firstly, the dispersion energy is defined in terms of the within-class distance and between-class distance. Secondly, with a kernel function, the dispersion information in local regions is extracted to establish the local dispersion-based active contour model, and a curve length energy term that weights by an edge indicator function is also incorporated into the novel model. Finally, a penalty term is added to avoid reinitializing periodically during the evolution of the level set method. Experimental results for both synthetic images and real images show desirable performances of the proposed method.