Computer Vision News - March 2017

26 Computer Vision News Research Research Selective Level Set Segmentation Using Fuzzy Region Competition Every month, Computer Vision News reviews a research from our field. This month we have chosen to review Selective Level Set Segmentation Using Fuzzy Region Competition , a research paper proposing a new level set formulation using fuzzy region competition for selective image segmentation. We are indebted to the authors ( Bing Nan Li, Jing Qin, Rong Wang, Meng Wang and Xuelong Li ) for their kind clarifications and for allowing us to use their images to illustrate this review. The research is here . Segmentation is one of the key tasks in image processing and computer vision. Many techniques have been suggested and used to perform this task, some of them with significant success: deformable models, active contour models and level set methods deservedly achieved major popularity among scientists and researchers. The main downside of edge-based level set methods is the necessity for objects to have a clear and distinct boundary, while in practical cases they are often found to be weak or discontinuous. Fuzzy logic techniques were introduced to cope with this issue of boundary leakage, with irregular success. This paper proposes a new level set model with fuzzy region competition for selective image segmentation; it also experiments (using Matlab) its advantages in selective level set segmentation on both synthetic and real images. Introduction: The classic level set equation is the following partial differential equation (PDE): where is the function defining the segmentation region, or more precisely the segmentation contour. The function receives negative values for areas within the segmentation, and positive values outside the segmentation. At the contour, the segmentation function receives the value 0. ∇ is the geometric gradient and F is the velocity field, which sets the rate of change of the contour of the segmentation. Segmentation by level set starts by setting an initial segmentation 0 , then iteratively refining it step-by-step until arriving at a final segmentation. Classic level set methods suffered from some inherent drawbacks in handling noise, ambiguity, and inhomogeneity, which extensive investigations of the equation, and attempts at improvement of both speed and precision, tried to remedy. In optimization functions of this sort there is always a trade-off between the local components and the global components + ∙ | | = 0 , ( , , = 0 = 0 ( ,