摘要
针对超声衍射层析成像传统采用的双线性插值法重建精度不高的问题,提出一种高精度的核卷积插值重建算法.首先,根据标准的sheep and Logan体模算出重建数据点的值,再选用最小二乘非均匀快速傅里叶变换(LS-NUFFT)算法里的核矩阵作为卷积核,并用此核矩阵将非笛卡儿分布的重建数据点插值到笛卡儿网格内,最后用二维的傅里叶逆变换完成图像的重建.与双线性插值法和高斯核卷积法相比较,LS-NUFFT核矩阵法所得重建图像的2-范数误差比双线性法减少了40%以上,重建时间比高斯核卷积法减少约50%.
To overcome the shortage of the conventional bilinear method in diffraction ultrasound tomography reconstruction, an accurate reconstruction algorithm is proposed by using kernel convolution interpolation. The first step of the algorithm calculates the reconstruction data by Shepp and Logan, and the kernel matrix obtained in the least squares nonuniform fast Fourier transform （LS-NUFFT） algorithm is used for the convolution kernel. Then the kernel matrix is used to interpolate the non-Cartesian sampling reconstruction data into the Cartesian grid. The image reconstruction is finally fulfilled by 2-dimensional IFFT. Comparisons with the bilinear method and the Gauss kernel convolution method show that, the reconstruction error in 2-norm is reduced more 40% than the bilinear method, and that the reconstruction time is shortened about 50% compared with the Gauss kernel convolution method.
出处
《西安交通大学学报》
EI
CAS
CSCD
北大核心
2009年第10期94-98,共5页
Journal of Xi'an Jiaotong University
基金
国家自然科学基金资助项目(60603083)
安徽省优秀青年人才基金资助项目(2008jq1156)
教育部高等学校博士学科点专项科研基金资助项目(20070357003)
关键词
衍射层析成像
非均匀快速傅里叶变换
核矩阵
diffraction tomography
nonuniform fast Fourier transform
kernel matrix
