岩心三维CT图像超分辨率重建  

Super-resolution reconstruction for three-dimensional core CT image

作　　者：张廷蓉 滕奇志[1] 李征骥[1] 卿粼波[1] 何小海[1] ZHANG Ting-rong, TENG Qi-zhi, LI Zheng-ji, QING Lin-bo, HE Xiao-hai(College of Electronics and Information Engineering, Sichuan University, Chengdu 610065, China)

机构地区：[1]四川大学电子信息学院,四川成都610065

出　　处：《浙江大学学报（工学版）》2018年第7期1294-1301,共8页Journal of Zhejiang University：Engineering Science

基　　金：国家自然科学基金资助项目(61372174)

摘　　要：为了提高岩心三维图像分辨率,将调整的锚点邻域回归算法(A+)扩展为三维图像超分辨率重建,提出三维高频修正A+算法.该算法利用已有的高分辨率(HR)岩心三维CT图像和高频修正信息训练高低分辨率字典、高频修正字典、映射矩阵和高频修正映射矩阵.重建时,对每个输入的三维低分辨率(LR)特征块搜索匹配的字典原子以及相应的映射矩阵和高频修正矩阵,通过LR特征向量分别与映射矩阵和高频映射矩阵相乘,直接将三维LR特征映射到HR空间.针对多组岩心三维CT图像进行实验,与其他三维超分辨率算法进行比较.实验结果表明,该算法具有较高的峰值信噪比和结构相似度.A ＋（adjusted anchored neighborhood regression）algorithm was extended to three-dimensional（3 D）image super-resolution reconstruction in order to improve the resolution of three-dimensional image of core.A 3 Dhigh frequency correction A ＋algorithm was proposed.The existing high resolution（HR）core 3 D CT image and the high frequency correction information were used to train the high and low resolution dictionary,the high frequency correction dictionary,the mapping matrix and the high frequency correction mapping matrix.In reconstruction,the matched dictionary atom and mapping matrix were searched for each input of the 3 Dlow-resolution（LR）feature block and mapped directly to the HR space via multiplying the mapping vectors by LR feature vectors,respectively.Several 3 Dcore CT images were experimented.The algorithm was compared with other three-dimensional super-resolution algorithms.The experimental results show that the proposed algorithm can obtain higher peak signal to noise ratio and structural similarity.

关 键 词：岩心三维CT图像 超分辨率重建 高频修正 字典训练 映射矩阵 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]