Poor local contrast is a nettlesome problem in biomedical image analysis. The multiscale method based on mathematical morphology has been successfully used in local contrast enhancement. However, the computational time is high. In this paper, the existing method is much improved by using few feature levels in an efficient way. The new algorithm is tested by experiments. Compared with the previous method, the results show that the proposed method is more effective, faster and less sensitive to noise. It is a fast and simple method for eliminating the non-uniformity of the intensities in images.