Shape Recognition by Clustering and Matching of Skeletons

We perform the task of shape recognition using a skeleton based method. Skeleton of the shape is considered as a free tree and is represented by a connectivity graph. Geometric features of the shape are captured using Radius function along the skeletal curve segments. Matching of the connectivity graphs based on their topologies and geometric features gives a distance measure for determining similarity or dissimilarity of the shapes. Then the distance measure is used for clustering and classification of the shapes by employing hierarchical clustering methods. Moreover, for each class, a median skeleton is computed and is located as the indicator of its related class. The resulted hierarchy of the shapes classes and their indicators are used for the task of shape recognition. This is performed for any given shape by a top-down traversing of the resulted hierarchy and matching with the indicators. We evaluate the proposed method by different shapes of silhouette datasets and we show how the method efficiently recognizes and classifies shapes.