Neutron Digital Image Restoration By Wavelet-based Contourlet Transform And RL

As a branch of nondestructive testing, neutron digital imaging is the technologyof acquiring the inner structure and detailed information of objects based on the factthat the ray is attenuated when penetrating into objects. Compared with conventionalnondestructive testing methods, such as the widely used X-ray imaging, neutrondigital imaging can detect the light materials obscured in heavy metals in themagnitude of millimeters, with great advantage in Hydride detection, and itsresolution increases by30~50times.Neutron digital imaging implements digital imaging in the system whereneutrons penetrate into the detected object in the neutron irradiation field. The digitalimage carries the information of an object’s inner structure. However, the imagegenerated by neutron digital imaging system is usually low contrast, blurred andpolluted by noise, which causes loss of detailed information of image and bringsnegative effects on analyzing the inner structure of the object. This paper attempts torestore the degraded digital image by denoise and deblur techniques.First the paper presents the main reasons of image degradation resulting fromneutron digital imaging, and concludes that the spatial blur caused by a variety ofnoise pollution and neutron scattering leads to neutron digital image degradation.Secondly, because of the limitations of traditional algorithms, the paperintroduces contourlet transform for image denoising. Contourlet transform iscomposed of two parts: wavelet transform and directional filter. Wavelet transformcan handle the problem of analyzing the singularity of one-dimensional signals, but itis very difficult to represent the geometric characters of high-dimensional signals,such as the contours, edges and features, which contain a large amount of informationof images. Contourlet transform can represent and process high-dimensional spatialdata of images efficiently, but it brings in much redundancy. So this paper introducescontourlet transform proposed by Eslami.Thirdly, deconvolution is utilized to address the problem of image blur, whichmainly caused by spatial blur resulting from neutron scattering. The point spreadfunction should be obtained exactly in traditional image restoration methods; however,the point spread function is very complex or even undetermined in the real neutrondigital images. To avoid the problem, the paper introduces Richardson-lucy algorithm, which can estimate the original image and point spread function by iterativecalculation.Finally, we realize our method under the environment of MATLAB, and theexperiments performed in neutron digital images show that the proposed methodachieves good results.