Natural interface for sketch recognition

© Springer-Verlag Berlin Heidelberg 2009. New interfaces for CAD applications are coming up due to new available devices endowed with alternate technology. The use of this kind of interfaces is not extended at all because of the lack of robustness and reliability in recognition and interpretation of user inputs. In this work we develop a first approach of an interpreter of user sketches suitable for working online in CAD environments. An agent based architecture for the recogniser is presented. The recognition process uses invariant descriptors as the fast Fourier transform of the radius and arc length versus cumulative turning angle signatures, the HU moments and other invariant shape factors. The final classification is carried out by statistical learning using non-linear discriminant analysis. The method proposed in this article is not dependent on the number of strokes, neither on the sketching sequence order powered by the user. Finally, a validation of the recogniser is done.