Abstract: This paper presents a novel approach for learning models from deformable shape contours using multi-scale matching.For the contours with a large degree of noise and deformation,firstly,contours are segmented and matched in multi-scale space.In this process,the scale parameter is adjusted according to the noise and deformation degree.Then,shape model is deduced from the correspondence of curve sequences. The proposed approach is effective to deal with noise and deformation of the models of closed contours.An application is reported to build the models of pedestrian contours with good result.