|Ph.D Student||Cohen Shira|
|Subject||Distortion Evaluation in Quantitative Radiography|
|Department||Department of Quality Assurance and Reliability||Supervisor||Professor Emeritus Amos Notea|
|Full Thesis text - in Hebrew|
Extraction of quantitative information from a radiographic image, and definition of edges or boundaries, is the main problem in a product analysis. The difficulty stems from distortion (blurring) of the image, induced by several factors including secondary radiation which is inherent to the process and uncontrollable.
The current accepted approach in extracting quantitative information from a digital radiographic image is based on image restoration (solving the "inverse problem") with the aid of deconvolution algorithms in the image plane, or spatial frequency filtering.
Processing of a radiographic image involves a characteristic spread function [Line Spread Function (LSF) in 1D or Point Spread Function (PSF) in 2D] which covers the totality of the above distortion factors.
In literature, the characteristic spread function is usually considered as symmetric and shift-invariant in the image plane, and the radiographic system is displayed as linear and isoplanar. However, these assumptions are approximations and in many cases inapplicable - for example, where density and atomic number vary from region to region in the image plane due to material combinations or mixtures, so that the characteristic spread function also varies within the image and becomes shift-variant. Moreover, the 1D LSF, which is derived from a derivative of the response to a step change and usually described as normal or exponentially symmetric, can in certain situations have an asymmetric shape, depending strongly on the relative step-to-block heights. In such cases, a shift-invariant function is necessarily limited to the region of the change, and its extension to the whole image would lead to distorted reconstruction. No solution has been found to-date to these two difficulties.
In an attempt to resolve this problem, LSF was examined in the present research simulatively, theoretically and experimentally. The possibility of its asymmetry and shift-variance was confirmed.
A "saturation thickness" (ST) concept was established as a "workable", albeit approximate, solution for image processing, such that above it LSF is taken to be almost symmetric.
A novel methodology was devised, based on the "Fuzzy Logic" tool, for reconstruction of an image when LSF is shift-variant. This methodology permits rapid approximation based on a limited number of representative LSF functions, from which an optimal "weighted" LSF is composed for processing a desired region in the image.
The developed new approach was demonstrated for different shaped sliced objects and nuclear reactor fuel element.