An Image Denoising Method for Real Scene Based on Pixel-Level Noise Estimation

The majority of image denoising algorithms assume that the noise is evenly distributed white Gaussian noise, however, the image noise which collected in real scenes is more complex. In this paper, we propose a real image denoising method based on pixel-level noise estimation. The method is improved on the basis of the block-matching and 3D filtering (BM3D) image denoising algorithm. The noise estimation algorithm further introduces pixel-level non-local self similarity (NSS) on the basis of patch-level NSS prior. After detecting the flatness of the image block, the relevant parameters are adaptively adjusted, finally the noise estimation algorithm and the image denoising algorithm are combined block by block. Experiments show that this noise estimation method greatly reduce the required processing time while ensuring the accuracy of noise estimation. The image denoising effect has certain superiority compared with other classic traditional denoising algorithms.