Abstract—In this paper, a novel global noise reduction approach based on the sparse representation and nonlocal means algorithm is proposed to enhance the image qualities of various medical imaging modalities, including ultrasound images and magnetic resonance images. By using an overcomplete dictionary, a medical image is decomposed into a sparsest coefficients matrix populated primarily with zeros. A nonlocal means algorithm is developed to deal with these sparse coefficients to exploit the repetitive characters of structures in the whole image, realizing a “truly” global denoising. With synthetic and clinical data of ultrasound images and magnetic resonance images, this approach has been compared with other five state-of-the-art denoising methods. The experiments quantitative results demonstrate the effectiveness of our approach, especially superior in reducing the noise while well preserving the tissue details. It is concluded that our proposed approach is capable of enhancing image quality in both ultrasound and magnetic resonance images. It has a broad field of applications and will increase the diagnostic potential of the medical images.