Hand Drawn Symbol Recognition by Blurred Shape Model Descriptor and a Multiclass Classifier

In the document analysis field, the recognition of handwriting symbols is a difficult task because of the distortions due to hand drawings and the different writer styles. In this paper, we propose the Blurred Shape Model to describe handwritten symbols, and the use of Adaboost in an Error Correcting Codes framework to deal with multi-class categorization handwriting problems. It is a robust approach tolerant to the distortions and variability typically found in handwritten documents. This approach has been evaluated with the public GREC2005 database and an architectural symbol database extracted from a sketching interface, reaching high recognition rates compared with the state-of-the-art approaches.