Incremental on-line sketchy shape recognition with dynamic modeling and relevance feedback

An original principle of an incremental online sketchy shape recognition has been exploited. The proposed method identifies the sketch recognition process into two user-centered issues: users' drawing habits and input intensions. To capture the users' habit of sketching, a dynamic user modeling is adopted to build the user models in an incremental decision tree for each specific user. It can be used to recognize the 'possible shapes' dynamically by means of fuzzy matching based on the visiting frequency of each recorded shape of user models. To capture the user's input intensions, a relevance feedback method, which is widely used in multimedia retrieval, is introduced into the sketch recognition for the first time. It can adjust iteratively the weights of the similarity-based matching of the sketchy shape and refine the recognition results incrementally by the users' explicit feedback. Experiments have proved that the proposed method is both effective and efficient.