For reducing noise and preferably keeping the edge of image, the mean shift algorithm based on the HIS space is proposed. Firstly the image is changed from RGB space to HIS space. Because the correlation of HIS is very little, H,I and S separately is smoothed based on the different step sizes. Finally the smoothed image is gotten by transforming the HIS space into the RGB space. Because the correlation of the RGB space is more strong, the drawback of the image which could not be separately smoothed by R G and B is overcame. The convergence of the moving dot is strictly proved. We prove that the algorithm could get better result than Wiener filtering and the mean-shift based on RGB space by experiment.