Image denoising is a lively research field now. For solving this problem, non-linear filters based methods are the classical approach. These methods are based on local analysis of pixels with a moving window in spatial domain, but also have some shortcoming. Recently, because of the properties of Wavelet transform, this research has been focused on the wavelet domain. Compared to the classical nonlinear filters, the global multi-scale analysis characteristic of Wavelet is better for image denoising. So this paper proposed a new approach to use orthonormal Wavelet transform and distance constraint to solve this. Here, by minimizing the Stein's unbiased risk estimate (SURE) method to calculate the low frequency sub-band images for estimating. And convert the high frequency sub-band images to feature space, then use distance constraint to denoise by trained samples set. The experimental results show that the proposed method is efficiency and keep the detail ideally.