The image stitching based on feature point is not only invulnerable to the light,rotation and other factors,but also helpful to increase efficiency. It has become one of the most attended methods. Scale invariant feature transform (SIFT) is a better coner extraction algorithm,but there are still some problems in the matching step. A new matching principle based on system similarity theory was proposed,and then it was compared with the traditional matching principle. The problems with tradtional matching principle was pointed out,and also the precision and speed of the new principle was tested ,and the reason why the new principle get higher precision was analysed. By the new principle the matching features were detected successfully. And then,the transform matrix was calculated by the progressive sample consistency algorithm. Further more,it was optimized by the Levenberg-Marquardt (L-M) algorithm,and finally automatic image stitching was realized . At last,the stiching result of panorama camera image was given,further discussion about the new matching principle was also given in the end of the paper.