Gaussian noise is the main noise in flash X-ray radiographs, which will be magnified in the subsequent density inversion and other processing, and seriously affect the results of density reconstruction and object boundary extraction. Therefore, eliminating Gaussian noise is the key content of flash X-ray radiograph denoising research. According to the characteristics of image noise in flash X-ray radiographs and the rotational symmetry of object axes, this paper studies the denoising algorithm of flash X-ray radiographs based on BM3D (Block Matching and 3D Filtering). To overcome the defect that it is difficult to obtain high-quality similar blocks in flash X-ray radiographs, the image sources for providing similar blocks are increased by rotating and mirroring the noisy degraded images without destroying the noise independence. At the same time, by introducing grayscale transformation of image blocks, the grayscale value requirements in the original similarity requirements are reduced, the shape similarity requirements are improved, and the ability to obtain high-quality similar blocks is increased. Image denoising results show that the improved BM3D method in this paper achieves better denoising effect because the quality of similar blocks is guaranteed.