A first approach of pattern recognition using Generalized Fourier descriptors and SVM Johel Miteran miteranj@u-bourgogne.fr Abstract Fourier Descriptors are commonly used for binary pattern recognition. They are invariant under translation, rotation and change of the perimeter of the boundary. Considering the group of motions in the plane, Gauthier et al. ( 1991 ) proposed a family of invariants, called Motion Descriptors, which remain unchanged under motions of objects in bidimensional gray-level images. Fonga (1996) extented these descriptors and proposed and algorithm for real time computation. We propose here to use this family of descriptors in the RGB space, combined with a supervised classification method such as SVM. During the training, a video sequence of the object is analyzed and the descriptors are computed for each image, allowing 3D object recognition. Since the computation is very fast, many applications can be studied (image retrieval, face recognition, part of matching process, etc).