Curvelet-based Bilinear Interpolation Method for Low-dose CT
Bo Meng, Huiqin Jiang, Zhanwei Liu, Zhongyong Wang, Yumin Liu
School of Information Engineering and Digital Medical Image Technique Research Center,
Zhengzhou University, Zhengzhou, China, 450001
iehqjiang@zzu.edu.cn
ABSTRACT
In this paper, a curvelet-based noise suppression bilinear interpolation method for low-dose CT images is proposed.
Curvelets provide a multidirectional and multiscale decomposition that has been mathematically shown to represent
distributed discontinuities such as edges better than traditional wavelets. Because the traditional linear interpolation
results in boundary fuzziness in interpolated images, combined with the advantages of curvelet transform, here we
propose a curvelet-based modified bilinear interpolation to improve the accuracy of interpolation. Extensive experiments
indicate that the proposed method can effectively improve the quality of the obtained target image based on low-dose CT
images and the produced slice image is similar to original slice image.
Keywords: Low-dose CT, Curvelet Transform, Bilinear interpolation, Cycle spinning.
1. INTRODUCTION
Low-dose CT scanning results in increased quantum noise and reduced spatial resolution of volume data
[1]
. In order
to improve the quality of 3D reconstructed images based on low-dose CT images, it is necessary to study the
cross-sectional interpolation. Classical image interpolation methods fall into three main categories: gray-based,
shape-based and wavelet-based interpolation. Gray-based interpolation methods involve linear, spline, and so on. The
gray-based method such as linear interpolation is used widely because it can be easily calculated. However, this kind of
interpolation results in boundary fuzziness. To overcome the drawback, Herman et al.
[2]
proposed shape-based methods
by encoding the segmented image with distance codes. This approach can maintain better geometric characteristics by
interpolating the distance instead of the gray values. However, it cannot deal effectively with objects with holes or large
offsets. Recently, Jian Wu
[3]
proposed a novel shape-based interpolation algorithm. This method can successfully resolve
the problem of boundary fuzziness. But it is sensitive to noise and loses some useful information. In order to suppress
noise in interpolation, a wavelet-based interpolation
[4]
has been developed. However, wavelet in 2-D images is only
better at isolating the discontinuities at edge points but cannot detect the smoothness along the edges
[5]
.
In this paper, we propose a curvelet-based noise suppression bilinear interpolation method for low-dose CT images.
The major contribution of this method is to combine a linear interpolation and a modified bilinear interpolation in the
curvelet transform domain. Experimental results show that the proposed method not only can be better to keep detail
information such as boundaries, but also can effectively suppress the noise.
2. METHODS
2.1 Curvelet Transform
The curvelet transform is a multiscale geometrical transform with frame elements given by scale, location and
orientation parameters. It has not only the time-frequency localization properties of wavelet but also shows a very high
degree of directionality. Curvelet transform can sparsely characterize the high-dimensional signals, it means the balance
between parsimony and accuracy will be much more favorable and a lower MSE results. In this paper, we choose the
wrapping-based Fast Discrete Curvelet transform (FDCT) because this transform is more intuitive and faster. The
curvelet decomposition yields six scale layer coefficients. The most inner layer is called the coarse scale layer. The
outermost is known as fine scale layer. The middle four layers is called detail scale layer. We reconstruct the single-layer
curvelet coefficients as illustrated in Fig.1. Where the coarse scale denotes most of the energy of the original image and
denotes the low-frequency estimation after the fast discrete curvelet transform. The fine scale and the detail scale denote
the high frequency coefficients which reflect the detail information and edge features.
Fifth International Conference on Digital Image Processing (ICDIP 2013), edited by Yulin Wang, Xie Yi, Proc. of SPIE
Vol. 8878, 88783X · © 2013 SPIE · CCC code: 0277-786X/13/$18 · doi: 10.1117/12.2030550
Proc. of SPIE Vol. 8878 88783X-1
Downloaded From: http://proceedings.spiedigitallibrary.org/ on 01/08/2014 Terms of Use: http://spiedl.org/terms