Since traditional discrete orthonormal wavelet transform being more sensitive to the original position of signal.In the paper,we propose a stationary wavelet transform which possess the properties of rotation,shift and scale invariance.First transforming the image to polar coordinates,and confirming the principal axis orientation to adopt orientation energy function,then rotating the principal axis of image to the level orientation to gain the unifying image in the orientation.By reforming the image and rotation,shift and invariance to the wavelet bases,thus we can eliminate the influence of shift and scale.In the paper,we use Bayesian adaptive threshold estimation to gain the approximate optimization de-noising threshold.The image possesses the most PSNR.Thus we propose a kind of method which uses the input data to gain the asymptotic optimization de-noising threshold.Finally,experimental result show that the method can better reserve the image edge and de-noise at the same time,and being a kind of valid method of image de-noising.