Feature Preserving Smoothing of Shapes Using Saliency Skeletons

We present a novel method that uses shape skeletons, and associated quantities, for feature-preserving smoothing of shapes in digital images. We preserve, or smooth out, features based on a saliency measure that relates feature size to local object size, both computed using the shape’s skeleton. Low-saliency convex features (cusps) are smoothed out, and low-saliency concave features (dents) are filled in, respectively, by inflating simplified versions of the shape’s foreground and background skeletons. The method is simple to implement, works in real time, and robustly removes large-scale contour and binary speckle noise whereas preserving salient features. We demonstrate the method with several examples on datasets the shape analysis application domain.