7th World Congress on Industrial Process Tomography
Multi-core CPU based three-dimensional image reconstruction for limited view tomography
M. Bajpaia, P. Guptab, P. Munshia
aNuclear Engineering & Technology Programme, Indian Institute of Technology Kanpur, Kanpur-208016, India
bDepartment of Computer Science & Engineering, Indian Institute of Technology Kanpur, Kanpur-208016, India
Abstract
We propose a block-Iterative MART algorithm and its implementation on a multi-core CPU system to explore the task parallelism nature of linear algebra based tomographic inversion for 3D image reconstruction from limited view data. The block iterative nature of the algorithm reduces the memory requirement for 3D image reconstruction and its parallel implementation gives the results in reasonable time. The results indicate that we can perform single iteration of block-Iterative MART algorithm in 1506 seconds for 512x512x512 volume size with 200 projections over a span of 3600. We have used 16 processors for this study.
Keywords: Block-Iterative MART, 3D image reconstruction, Cone beam Computed Tomography, Parallel algorithm
Sign-in to access the full text
Copyright © International Society for Industrial Process Tomography, 2013. All rights reserved.