Skip to Main Content
Image segmentation is the critical task of partitioning an image into multiple objects. Deformable Models are effective tools aimed at performing image segmentation. Among them, Topological Active Nets (TANs), and their extension, ETANs, are models integrating features of region-based and boundary-based segmentation techniques. Since the deformation of the meshes composing these models to fit the objects to be segmented is controlled by an energy functional, the segmentation task is tackled as a numerical optimization problem. Despite their good performance, the existing ETAN optimization method (based on a local search) can lead to result inaccuracies, that is, local optima in the sense of optimization. This paper introduces a novel optimization approach by embedding ETANs in a global search memetic framework, Scatter Search, thus considering multiple alternatives in the segmentation process using a very small solution population. With the aim of improving the accuracy of the segmentation results in a reasonable processing time, we introduce a global search-suitable internal energy term, a diversity function, a frequency memory population generator and two proper solution combination operators. In particular, these operators are effective in coalescing multiple meshes, a task previous global search methods for TAN optimization failed to accomplish. The proposal has been tested on a mix of 20 synthetic and real medical images with different segmentation difficulties. Its performance has been compared with two ETAN optimization approaches (the original local search and a new multi-start local search) as well as with the state-of-the-art memetic proposal for classical TAN optimization based on differential evolution. Our new method significantly outperformed the other three for the given set of images in terms of four standard segmentation metrics.