A Project for Hand Gesture Recognition CARLOS R. P. DIONISIO ,ROBERTO M. CESAR J R Department of Computer Science, Univ. of S˜ ao Paulo - Caixa Postal 66.281, 05315-970 - S˜ ao Paulo, SP, Brasil lobo,cesar @ime.usp.br Abstract. This work discusses an ongoing project for hands gesture recognition in computer vision systems. The proposed approach is based on the shape analysis tools introduced in [1]. More specifically, the wavelet transform will be used for generating features for the recognition of hand gestures based on contours. 1 Introduction Hand gesture is an important way for communication be- tween people. On a simple human-machine interface, the hand gesture recognition is very important. Possible appli- cations include from console replacement to the communi- cation with a deaf as a virtual reality device. In the devel- opment of systems based on gesture recognition, there are three problems: Static Recognition of Hands Form; Hand Tracking; Dynamic Gesture Recognition. Firstly, we are interested in the static recognition of hands form. R. Cesar [1] on his doctoral thesis, introduces a new approach for object recognition based on contour anal- ysis. This work discusses an application of this new ap- proach for hand gesture recognition. 2 Multi-Scale Analysis of Two-Dimensional Forms R. Cesar [1] introduces conceptual ideas and algorithms for representation and multi-scale analysis of shape contours in digital images. That work also presents several techniques for contour analysis by multi-scale curvature and Gabor transform including specific algorithms for corner detec- tion, natural scale characterization, fractal analysis of self- similar curves and feature vector extraction associated with different shape aspects such as complexity and rectangu- larity. Figure 1 shows the contour of a hand gesture, while Figure 2 shows the corresponding continuous wavelet trans- form using the mexican hat wavelet. Our ongoing work deals with defining and extracting feature vectors from this type of representation, as well as the application of auto- matic feature selection algorithms for the design of more robust classifiers. It is worth noting that an analogous ap- proach has been successfully applied for the recognition of neural cells in [2], thus paving the way for the development of the present project. Figure 1: Contour of a hand gesture, adapted from the data made available in [3]. Figure 2: Continuous wavelet transform using the mexican hat wavelet. Acknowledgments Roberto M. Cesar Junior is grateful to FAPESP for the fi- nancial support (98/07722-0 and 99/12765-2), to “pro-reitoria de pesquisa” and to “pro-reitoria de p´ os-graduac ¸˜ ao” - USP, as well as to CNPq (300722/ 98-2). Carlos is grateful to Capes. References [1] R.M. Cesar Jr. and L. da F. Costa, An´ alise Multi-Escala de Formas Bidimensionais, Institute of Physic of S˜ ao Carlos, Univ. of S˜ ao Paulo (S˜ ao Paulo, 1997). [2] R.M. Cesar Jr. and L. da F. Costa, Neural Cell Classifi- cation by Wavelets and Multiscale Curvature, Biologi- cal Cybernetics, 79(4):347-360, 1998. [3] E. Milios and E.G.M. Petrakis, Shape Retrieval Based on Dynamic Programming, IEEE Transactions on Im- age Processing, vol. 9, no. 1, 2000, 141-146. Proceedings of the XIII Brazilian Symposium on Computer Graphics and Image Processing (SIBGRAPI00) 0-7695-0878-2/00 $10.00 ' 2000 IEEE