Study On Image Processing Techniques In Neutron Digital Imaging

Modern digital imaging can not go without digital image processing. Neutron digital image, obtained from the neutron digital imaging system after neutrons penetrate through the sample, carries information of the sample's internal structure. Unavoidably affected byγ-ray interference, neutron scattering and CCD's electronic noise, neutron images always suffers from degradation, e.g. corrupted by intense noise, detail-blurred, thereby it is not advantageous to analyze and determine the information about the sample's internal structure.Therefore, the de-noising and de-blurring method of neutron digital images is discussed, which belongs to the category of image processing techniques, so as to get a clearly insight into the object's internal structure and enrich the content of neutron non-destructive testing technique. According to the theory of neutron radiography and neutron imaging, the main reason for degradation of digital neuron image is the corruptions caused by various noises and spatial blurring, therefore the research work is carried out on this basis.In the research concerning de-noising, discrete wavelet transform (DWT) is introduced instead of some traditional de-noising methods that have limited usefulness. Focusing on the wavelet domain, for various random white noises, firstly, the statistic characteristics in wavelet domain and traits in different DWT scales are examined; owing to the characteristic of DWT, steady random white noise could be regarded as Gauss-distribution random white noise and filtered. Thus, wavelet threshold de-noising methods is chosen, also analysis and comparisons are made among some of them; finally one proper threshold type is specified. These works still lay groundwork for de-convolution when the noise level of neutron digital image is comparatively high.In the research concerning de-blurring, since spatial blurring is caused by point expanding, the de-convolution is considered as the key point to solve this problem. First some typical de-convolution methods for de-blurring are examined, such as inverse filtering, Wiener filtering restoration. A point spread function is necessary while applying these methods, whereas the one of neutron digital images is complex and usually unknown so that these methods can barely work. Under the circumstances, Richardson-Lucy algorithm is brought in and simultaneously the estimation of the original image and that of the point spread function is proceeded by alternate iteration...