Rapid processing of CCD star image is a key technique for star sensors as well as other optic surveillance sensors. The foreground of CCD star image is different from the background distinctly. A method of star image de-noising based on Top-hat transformation and median filter is presented. This de-noising method is able to hold the edge of stars. Then a new algorithm based on grads Hough transformation for star point rapid locating is presented. The new algorithm is simpler and high efficiency for star image processing. Experiments on real CCD star image with size of 752×580 indicats the validity of the algorithm.