Mihcak et al. proposed a low complexity but powerful image denoising algorithm LAWML based on the decimated wavelet transform (DWT). The shortcoming of LAWML is to determine the global optimal neighboring window size by experimenting. We improve on LAWML using Stein's unbiased risk estimate(SURE). Our method can automatically estimate an optimal neighboring window for every wavelet subband. Its denoising performance also surpasses LAWML because the subband adaptive window is superior to the global window. Furthermore, our method on the DWT is extended to on the dual-tree complex wavelet transform (DT-CWT). Experimental results indicate that our method (DT-CWT) delivers the comparable or better performance than some of the already published state-of-the-art denoising algorithms.