A new image denoising method combined a Pyramidal Dual-tree Complex Directional Filter Bank(PDTDFB) domain Gaussian Scale Mixture(GSM) model and a non-local mean filter was proposed. First, the locally coefficients PDTDFB(GSM) model for a noisy image was established, and the denoised coefficients were estimated by the Bayes least square estimator. Then, the inverse PDTDFB transform was used to obtain the preliminary denoised image. Finally, Nonlocal Mean Filter(NLMF) was employed to smooth the artifacts of the preliminary denoised image and to obtain the final denoised image. This method combines the characters of the PDTDFB on shift-invariance, multi-directional selectivity, image edge representation and the effective ability of GSM model for capturing correlation of neighbor coefficients. Experimental results indicate that the proposed method has removed Gaussian white noise while effectively preserving edges and texture information. Comparing with some outstanding denoised methods, its Peak Signal to Noise Radio(PSNR) value increases 0.3-3 dB and visual quality is obviously improved.