Low-dose CT not only reduces the radiation to human body, but also affects the image quality. Aiming at the problem of large noise in low-dose CT images, BM3D algorithm is applied to the denoising process of CT images in this paper. Based on the idea of non-local block matching, the algorithm firstly looks for image similarity blocks through block matching, then stacks the im- age similarity blocks into 3D matrix and performs collaborative filtering processing, and then aggregates the processing results into the original image blocks to restore the image. Compared with the traditional denoising algorithms, BM3D algorithm has great ad- vantages, better visual effect and significantly higher PSNR value.