Image Reconstruction from Limited-Angle Data Sets

• Main Authors: Tsai Chang-Han, 蔡昌翰
• Other Authors: Hung Hsien-Sen
• Format: Others
• Language: zh-TW
• Published: 1999
• Online Access: http://ndltd.ncl.edu.tw/handle/72794854240341933866

碩士 === 國立海洋大學 === 電機工程學系 === 87 === Reconstruction of cross-section images from the projections of an object is a widely used image processing technique. Traditional application of image reconstruction is the X-ray computed tomography for medical imaging, which reconstructs cross sections from projections of human body through the process of computing devices. In recent years, computed tomography has found its success in various applications, such as electron microscopy, astronomy, nondestructive evaluation, and many others. However, in many cases it is not possible to collect projection data over a complete angular range of. This is the so-called limited-angle problem that is mainly caused by the size of the object under test. Lack of complete angular coverge in CT scanning renders most of the Fourier-based image reconstruction methods, such as filtered back-projection (FBP), ineffective. As a result, they usually produce severe artifacts and also degrade accuracy in reconstructed cross sections. The iterative reconstruction-reprojection (IRR) algorithm proposed by Medoff et al. is commonly employed to solve the limited-angle problem. However, lack of sufficient prior information makes IRR less effective in the performance improvement of reconstructed images. Besides, the IRR algorithm has slow convergence rate in a recursive fashion to regularize the limited-angle problem. Therefore, how to maximize the use of prior and accelerate the convergence of the IRR algorithm is the main goal of the thesis. To improve the performance of the IRR algorithm, flawless prototype image is incorporated and difference constraint is developed as additional constraints of prior information. In addition, the constraint in frequency domain is also incorporated to increase convergence rate. Thus the performance of the IRR algorithm in effectiveness and efficiency can be greatly improved.