In this work some post-acquisition image enhancement methods are applied to the case of under-sampled and low-resolution neutron radiography image. The digital neutron imaging system used in this study is a low resolved one; mainly based on an 8 bits CCD camera and a 250μm LiF-ZnS scintillator screen. In this paper, a software solution is proposed to improve the quality of digital neutron radiography images. The suggested method is a combination between the convolution neural network single-image and the projection onto convex sets multi-image super-resolution algorithms (SRCNN-POCS). The results obtained are compared to those by each algorithm separately. The proposed SRCNN-POCS method was found to be more reliable by producing higher quality image than the others tested methods in terms of modulation transfer function, peak-signal-to-noise ratio (PSNR), and structural similarity (SSIM). The visual analysis of the obtained image confirms too the obtained results.
Read full abstract