基于稀疏处理的多能X射线分离成像  

Separation of multi-energy X-ray imaging based on sparse processing

作　　者：费彬[1] 孙京阳 张俊举[2] 喻春雨[1] 

机构地区：[1]南京邮电大学光电工程学院,江苏南京210023 [2]南京理工大学电子工程与光电技术学院,江苏南京210094

出　　处：《光学精密工程》2017年第4期1106-1111,共6页Optics and Precision Engineering

基　　金：高等学校博士学科点专项科研基金资助项目(20133223120007);江苏省自然科学基金(BK20140876)

摘　　要：利用独立成分分析(Independent Component Analysis,ICA)并结合多能X射线图像的丰富信息可以将二维X射线图像中重叠目标分离成像,但是海量的图像数量,以及高像素数的要求均会使内存占有量和计算速度面临挑战,因此本研究将压缩感知(Compressed Sensing,CS)与ICA相结合进行分离成像,以提高计算速度和分离成像性能。研究过程中,首先根据被拍摄物体的物质组成确定拍摄多能X射线图像数量,并选取CS技术中K均值奇异值分解(K-means SingularValue Decomposition,K-SVD)稀疏基将多能X射线图像进行稀疏表示,然后利用ICA将此稀疏表示进行盲源分离得到独立源,最后采用正交匹配追踪算法(Orthogonal Matching Pursuit,OMP)将独立源进行重构实现分离成像。研究结果表明:采用ICA&CS技术比仅采用ICA进行目标分离成像的运行时间减少了46.14s(23.3%)、内存占有率降低了21%、重构图像峰值信噪比(Peak Signal to Noise Ratio,PSNR)提高了2.665dB、边缘梯度提高了0.001、信息熵提高了0.09。Independent Component Analysis （ICA） combined with abundant information of multi-ener- gy X-ray images can achieve the imaging separation of overlapping targets in 2D X-ray images, but the increasing number of images and higher pixel requirements may serve as an obstacle for memory occu- pancy and calculating speed. In this paper, Compressed Sensing （CS） was combined with ICA to a- chieve the imaging separation and to improve the calculating speed, as well as the imaging separation performance. First, the number of the multi-energy X-ray images was determined based on composi- tion of the captured object, and then sparse representation of multi-energy X-ray images was carried out by selecting K-means Singular Value Decomposition （K-SVD） in the CS technology then, Blind Source Separation（BSS） was conducted in such sparse representation to obtain the independent source by using ICA finally, Orthogonal Matching Pursuit （OMP） was used to reconstruct the independentsource, thus achieving the imaging separation. The results show that compared with the algorithm merely based on ICA, ICAICS could reduce the algorithm running time by 46. 14 s （23.3%） and memory occupancy by 21%, and improve the Peak Signal to Noise Ratio （PSNR） of the reconstructed image by 2. 665 dB, edge gradient by 0. 001 and information entropy by 0.09.

关 键 词：压缩感知 独立成分分析 多能量成像 图像重构 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]