Abstract: A BP neural network based image restoration algorithm is proposed against the blur of flash radiographic images with low signal-to-noise ratios. Firstly the degraded image and the sample image are both divided into the edge part and the smooth part. Secondly the nonlinear mapping relationship between the degraded image and its original image is established by training the BP neural network which has the ability of learning, remembrance and generalizing with the edge part of the sample image. Finally the edge part of the degraded image could be restored by the trained neural network. Numerical results show that this method could restore the image information near edges when the systematic point-spread function is unknown, and the restored image is greatly improved in the signal-to-noise ratio and in the visual.