In order to solve the computation increase caused by high order local features, this paper proposed a novel high order local feature representation approach based on kernel function. It first compared local features extracted from two difference images. Then transformed the features in the two images to the geometry invariant spaces. Next, it constructed kernel function by counting high order features. Finally, it performed the multi object classification experiment by using support vector machine. The experimental results show that proposed method has better precision and its computation time is linear to the number of local features in images.