We have proposed a new implementation of contourlet, namely, WDFB contourlet which is very efficient for image denoising. We also find two reliable methods for norm estimation, i.e. Monte-Carlo estimation and polyphase impulse reconstruction. The former is fast, simple yet reliable, while the latter can exactly reconstruct the contourlet basis, thus lead to precise results. Then shearing invariant (SI) is proposed as a novel tool for further denoising improvement, especially for multiscale directional transforms. Finally, numerical experiments are presented to support our conclusions.