Abstract: In high-energy transmission industry computed tomography (ICT),
convolution back-projection (CBP) is widely used for image reconstruction. But
high mass thickness and the requirement of short scanning time lead to high
statistical noise. Statistical methods are developed from emission tomography
and its potential in transmission tomography has been appreciated recently.
Consequently, statistical methods, which model the Possion noise of X-ray
photons, are expected to produce better images than transform methods. In order
to use those methods, such as maximum a posteriori (MAP), we need speed up the
iterate algorithm and make use of the work-piece's prior information
thoroughly. This paper presents a novel prior image model, which we refer to as
genetic discrete Markov random filed (GDMRF) model. This model has two useful
characters· discrete gray set and the genetic character of gray set.
With the discrete gray set, the iterate algorithm is highly active. With the
genetic character of GDMRF model, the reconstructed image quality benefits from
using prior information thoroughly. Experiment has demonstrated that the GDMRF
model is useful for image reconstruction from sinogram with low signal-to-noise
ratio (SNR).
