COMPRESSIVE SENSING BASED IMAGE RECONSTRUCTION FOR COMPUTED TOMOGRAPHY DOSE REDUCTION

Excessive radiation exposure is one of the major concerns in the computed tomography (CT) field. Few-view reconstruction using iterative algorithm is an important strategy to reduce the radiation dose. In the iterative CT reconstruction, the projection / backprojection model plays an important role in the overall computational cost, image quality, and reconstruction accuracy. In this dissertation, we first propose an improved distance-driven model (IDDM) whose computational cost is as low as the well-known distance-driven model (DDM) and the accuracy is comparable to the accurate area integral model (AIM). Recently, the Lp (0<p<1) regularization has attracted a great attention because it can generate sparser solutions than the L1 regularization. We derive several analytic thresholding representations for Lp (0≤p≤1) regularization and develop a corresponding general thresholding algorithm which is adequate and efficient for large-scale problems such as CT reconstruction. The analytic thresholding representation for the Lp regularization permits a fast solution similar to the iterative hard thresholding algorithm for the L0 regularization and the iterative soft thresholding algorithm for the L1 regularization. The Lp (0<p<1) regularization is very sensitive to noise and the initialization has a significant influence on the performance. We finally propose an alternating iteration algorithm based on the derived analytic thresholding representations. For the proposed alternating iteration algorithm, the zero initialization works equally well as the L1 initialization and is robust to noise.

Grantor dc:publisher

Wake Forest University

Year dc:date.issued

2015

Author dc:creator

• Miao, Chuang

• Compress Sensing

Language dc:language.iso

en

Handle dc:identifier.uri

http://hdl.handle.net/10339/57252

OAI identifier oai:identifier

oai:wakespace.lib.wfu.edu:10339/57252