基于多景深融合模型的显微三维重建方法

发布时间：2019-06-21 08:32

【摘要】：鉴于激光共聚焦三维重建技术在金属样本的显微三维重建过程中具有重建速度慢的不足,在保证重建精度的前提下,提出一种基于多景深融合模型的显微三维重建方法来提升重建效率.该方法将三维重建问题转化为二维图像景深融合问题,首先提出一种基于非降采样Contourlet变换与多态脉冲耦合神经网络的图像融合方法,对不同景深图像序列运用该方法进行融合操作,得到的初代融合图像最大程度保留了原始图像信息;然后提出一种基于相关系数的区域图像匹配方法,得到以原始图像为基础的二次融合图像和初代高度映射图;最后定义一个能量泛函模型以消除二次融合图像中存在的假点信息,其极小化过程保证二次融合图像收敛于初代融合图像,同时初代高度映射图按照相同的方式演化,迭代结束即完成三维重建.运用文中方法对微尺度金属样本进行三维重建实验的结果表明,相比于激光共聚焦三维重建,该方法在保证三维重建结果具有较高精度与较好抗噪性的基础上,提升了重建效率.
[Abstract]:In view of the slow reconstruction speed of laser confocal 3D reconstruction in the process of microscopic 3D reconstruction of metal samples, a microscopic 3D reconstruction method based on multi-depth of field fusion model is proposed to improve the reconstruction efficiency on the premise of ensuring the reconstruction accuracy. In this method, the problem of 3D reconstruction is transformed into the problem of two-dimensional image depth-of-field fusion. Firstly, an image fusion method based on non-downsampling Contourlet transform and multistate pulse coupling neural network is proposed. This method is used to fuse different depth of field image sequences, and the original image information is preserved to the maximum extent. Then a region image matching method based on correlation coefficient is proposed, and the secondary fusion image and the primary height mapping map based on the original image are obtained. Finally, an energy functional model is defined to eliminate the false point information existing in the secondary fusion image. The process of minimizing the secondary fusion image ensures that the secondary fusion image converges to the primary fusion image. At the same time, the primary generation height map evolves in the same way, and the three-dimensional reconstruction is completed at the end of the iteration. The experimental results of three-dimensional reconstruction of microscale metal samples show that compared with laser confocal three-dimensional reconstruction, this method improves the reconstruction efficiency on the basis of ensuring that the three-dimensional reconstruction results have higher accuracy and better anti-noise.
【作者单位】：中国科学院成都计算机应用研究所机器视觉研究室;中国科学院大学;中国科学院广州电子技术研究所数理研究室;中国科学院广州地球化学研究所有机地球化学国家重点实验室;广州大学计算机科学与教育软件学院;
【基金】：国家“八六三”高技术研究发展计划(2015AA015408) 中国科学院西部之光基金(2011180)
【分类号】：TP391.41

【相似文献】