A correction algorithm of geometric unsharpness for neutron radiographs via adaptive fusion and total variation

Abstract

The mobility of accelerator-based neutron sources has laid the foundation for the miniaturization of neutron radiography facilities in response to the increasing industrial requirements. Compared to traditional reactors and spallation neutron sources, miniaturized neutron radiography facilities based on accelerator neutron sources face challenges in achieving a sufficiently high collimation ratio. It leads to a pronounced level of geometric unsharpness on neutron radiographs, which arises from a finite source size and beam divergence. The presence of blurriness in radiographs significantly impairs the ability to accurately visualize the internal structures of the test objects, thus hindering observation. An algorithm combining adaptive fusion with total variation was proposed to correct the geometric unsharpness for neutron radiographs in this paper. In contrast to most traditional restoration algorithms that relied on regularization terms or filters, our proposed method offered both initial restoration and further restoration. The initial restoration procedure involved the utilization of the Richardson-Lucy algorithm coupled with the l0 smoothing method. This adaptive fusion technique was employed to enhance the sharpness of edges and denoise the neutron radiographs. Finally, the result of the adaptive fusion was embedded into a deblurring model based on total variation to further preserve the edges and restore more details during further restoration. In the experiments, it was demonstrated that our correction algorithm effectively preserved the prominent edges and the details in both the simulated and real neutron radiographs of thin and thick objects, exhibiting fewer ringing effects than traditional algorithms. It is worth mentioning that our work is primarily applicable and significant to neutron generators, but not for reactors or spallation neutron sources that can deliver much better images.