The Research On Skeleton Computation Method For Two-Dimensional Shapes

Skeleton is an abstract of the shape,it denotes the topological structure and features of original shape,it is essential for shape representation and analysis in many application areas such as shape retrieval,animation deformation,object recognition,human pose recognition,circuit board inspection systems and analysis of biomedical images.The definition of the skeleton is proposed by Blum in 1967,which is the set of the centers of the inscribed balls,the skeleton point has at least two nearest points in the boundary.A common problem in skeleton computation is that little protrusion of the boundary often generates redundant skeleton branches that may seriously affect the correctness of the skeleton topology.Aiming at the problem that the skeleton computation method is sensitive to boundary noise,some stable skeleton extraction method has been proposed and some post-processing pruning methods are also used to trim off the abundant branches.The improvement of the skeletonization enhances the application of the skeleton in object detection in natural images and contour extraction,obtaining a skeleton that is stable and can represent the geometric features of the original object has important research significance and application value.Since Blum proposed the definition of the skeleton,many scholars have proposed a variety of skeleton computation methods,however,these methods have more or less defects.There is still no method to obtain the ideal skeleton result of a two-dimensional shape directly.We give the properties of the ideal skeleton and the ideal skeleton computation methods:(1)skeleton conformal to the input shape topologically,(2)skeleton computation method is not sensitive to the shape boundary noise,(3)skeleton computation doesn't require post-processing operations such as pruning,(4)skeleton computation doesn't require manual parameter assignment.In this thesis,for how to get an ideal stable skeleton for two-dimensional shapes,we propose a new method to extract two-dimensional skeleton based on heat equation.The algorithm contains the following steps:firstly,using the smoothness of the heat equation to smooth the distance field,diffusing from the edge of the shape to the inner,a smooth distance surface will be built,which is called heat surface.The heat surface keeps the features of the distance map and smooth the noise from the boundary,which effectively reduces the impact of boundary noise on the final skeleton.Secondly,extracting ridges from the heat surface,some ridge curves are reserved as a part of the skeleton.Finally,a skeleton completion is used to ensure the completeness,trace form the disconnected points,along the direction of the steepest descent to complete the skeleton,then the final skeleton is got.In experiments,we choose the datasets that contain kinds of scales of shapes to experiment.The experiments show,comparing with the representative skeletonization methods,more stable skeletons which can keep the feature of the original shapes can be got from the method proposed by us.The results of the shapes full of boundary noise performs better,the skeleton can still be simple,stable and good to express the original shapes,the reconstruction error ratios also prove our method's good performance.