Abstract: The neutron radiography is one of the most widely used techniques for non-destructive testing. However, the resulting images are degraded inevitably due to some physical limitations, such as noise corruption and blur. To solve this problem, a new scheme for neutron image restoration was proposed, which utilizes BM3D frames and variance stabilizing transformation. In this scheme, the transformation was applied to produce the data which can be treated as an additive Gaussian noise. After denoising the resulting data by BM3D frames, an inverse transformation was applied to the denoised signal to obtain the restoration result. This scheme is capable of suppressing strong Poisson-Gaussian noise while restoring details of the degraded neutron images. Experimental results show that the scheme improves the restoration quality efficiently and possesses a robust performance both quantitatively and visually.