The invention discloses a low-dimension characteristic model image processing method based on scale invariant feature transform. For aiming at solution in a key point direction, eight adjacent pixels of each pixel are utilized for solving according to a gradient solving method, and gradient solving by means of an upper point, a lower point, a left point and a right point according to a document [1] is abandoned. SIFT is a local characteristic of an image, and only local information in the image is not comprehensive enough, and therefore increase of some global characteristics on the extracted local characteristic is necessary. According to the method of the invention, a concept of moment of inertia in physics is utilized for adding the global characteristic on the image. Furthermore for performing redundancy removal on a descriptor, further DCT transform is performed on the extracted descriptor, and the information with relatively low information amount is filtered, thereby realizing a purpose of redundancy removal.