Improving Shape Context Matching for the Recognition of Sketched Symbols

In this paper we present an approach to recognize multi- stroke hand drawn symbols, which is invariant with respect to scaling and supports symbol recognition independently from the number and order of strokes. The approach is an adaptation of the algorithm proposed by Belongie et al. in 2002 to the case of sketched images. This is achieved by introducing a new step in which the original Shape Context point-to-point cost matrix is updated according to stroke re- lated information. The approach has been evaluated on a set of symbols from the Military Course of Action domain and the results show that the new recognizer outperforms the original one.