Abstract: At autonomous image guidance systems,the missile recognizes targets according to the saved physiographic pictures,and SIFT is outstanding in image matching but has relatively computational load.The novel algorithm builds the step pyramid(the zoser pyramid) representation by successively reducing the image size with combined interpolation and does repeatedly not convolve the initial image with Gaussian at each scale of scale space.Each sample point makes a comparison with its 24 neighbors to detect the local maximum and minimum for each scale.It can exactly recognize objects in photo maps even though addition of the noise,affine distortion,and change in illumination.The algorithm processes more quickly and needs less memory units at the real-time on-board DSP system.