Research of Image Denoising Algorithm Based on Non-local Priori Constraints

Traditional Non-Local Means makes use of the characteristic of white Gaussian noise which has zero mean. This method only uses weighed mean as an estimator to evaluate each pixel. However, weighed mean is not an appropriate estimator for other kinds of noise. Hence, median filter is combined with pre-filtering and Median Pre-Filtering Non-Local Means is given out, whose denoising performance is pretty well in terms of both whit Gaussian noise and salt & pepper noise. In addition, a large quantities of experiments are conducted and some advices on choosing optimal parameters are proposed. As for most images, our advice can help the algorithm achieve great performance.