The automatic recognition by gait has recently gained more and more interests as the unique performance to recognition people at distance. An appearance-based approach to improve the gait recognition is proposed. The vector data scanned from horizontal

vertical and diagonal direction of the outer contour of binarized silhouette of a walking person are chosen as the image feature. These temporal and spatial feature sequences are decomposed based on the discrete wavelet transformation(DWT) to reduce data dimensionality and filter the noise produced from the procedure of template extracting. Then the multi-class support vector machine(SVM) models are trained by the decomposed feature vectors. The gaits are classified by the trained SVM models. This algorithm is applied to a data-set including thirty individuals. Extensive experimental results demonstrate that the proposed algorithm performs at an encouraging recognition rate with 91% at relatively lower computational cost.