基于驱动力的Log-Demons算法及其在大形变图像配准中的应用

发布时间：2018-05-04 03:19

本文选题：图像配准 + 大形变图像　；参考：《华东师范大学》2017年硕士论文

【摘要】：大形变图像配准在计算机图像处理尤其医学图像处理中有重要的研究价值和应用意义。由于在配准过程中形变量较大,传统的Demons算法仅仅利用图像的梯度信息驱使像素朝梯度下降的方向扩散,使图像发生形变,导致在图像平坦或者边缘处梯度信息缺失的情况下很难完成配准。因此,本文提出了基于驱动力的Log-Demons算法,首先通过提取图像的结构张量作为约束,使配准过程保持更多的结构信息,其次利用基于描述子匹配获得的驱动力提高图像在大形变情况下的配准精度,避免算法配准陷入局部最优,最终构建了基于Log-Demons算法的大形变图像高精度配准模型。本文的主要工作包括:1)提出基于结构张量的Log-Demons算法:图像的结构张量能提取更多的局部结构信息,并且对外部光照变化不敏感。通过张量守恒准则将其融合进Log-Demons算法中来约束像素点的扩散,使其能保持良好的局部结构并获得更精确的形变场。2)提出基于驱动力的Log-Demons算法:为了能实现大形变图像高精度配准,本文提出将边缘点匹配算法获得的矢量位移作为驱动力,拉动图像边缘周围像素点的运动,使其朝正确的方向扩散,解决了大形变情况下图像边缘梯度相同导致像素点随机扩散的问题,使其能够适应大形变图像配准。3)提出基于MROGH描述子匹配获得驱动力的算法:为了获得更准确的驱动力,并适应大形变带来的大角度变化,本文提出采用MROGH构建描述子并对其进行一对一的精确匹配以获得匹配点的位移向量作为驱动力。4)提出基于特征驱动的Log-Demons融合算法:将驱动力作为常量单独计算,在迭代配准的过程中以指数减少的方式与Log-Demons的更新形变场相融合,使其在形变开始时给予较大的影响力,形变减小时由Log-Demons自身的驱动力完成配准,既加快配准过程又减少了特征点误匹配所带来的影响。在更新形变场计算过程中,本文提出在李群中利用群性质融合驱动力和Log-Demons本身的更新形变场,使融合后的形变场也能保持微分同胚。通过算法(1)-(4),最终构建了基于驱动力的Log-Demons算法模型,实现了大形变图像配准。本文方法在模拟形变图像、真实场景大位移图像以及脑图像上进行了测试,并与其它方法(如Log-Demons、SpectralLog-Demons、LDDMM)进行了比较,实验表明本文方法都取得了较好的效果。本文方法不仅能较好地完成大形变图像配准,而且计算得到的形变场具有更好的局部结构和精度。
[Abstract]:Large deformation image registration has important research value and application significance in computer image processing, especially in medical image processing. Because of the large amount of deformation in the registration process, the traditional Demons algorithm only uses the gradient information of the image to drive the pixel to the direction of gradient descent, which causes the image to deform. It is difficult to complete registration when the image is flat or the gradient information at the edge is missing. Therefore, a driving force based Log-Demons algorithm is proposed in this paper. Firstly, by extracting the structure Zhang Liang of the image as the constraint, the registration process can keep more structure information. Secondly, the driving force based on descriptor matching is used to improve the image registration accuracy in the case of large deformation, and avoid the local optimal registration. Finally, a high-precision registration model of large deformation image based on Log-Demons algorithm is constructed. The main work of this paper includes: 1) A Log-Demons algorithm based on structural Zhang Liang is proposed: the structure tensor of the image can extract more local structure information and is not sensitive to the external illumination change. The Zhang Liang conservation criterion is used to integrate it into the Log-Demons algorithm to constrain the spread of pixels. The Log-Demons algorithm based on driving force is proposed. In order to achieve high precision registration of large deformation image, the vector displacement obtained by edge point matching algorithm is used as the driving force in this paper. The motion of pixels around the edge of the image is pulled to spread in the right direction, which solves the problem of random diffusion of pixels caused by the same gradient of the edge of the image under the condition of large deformation. So that it can adapt to large deformation image registration. 3) an algorithm for obtaining driving force based on MROGH descriptor matching is proposed: in order to obtain more accurate driving force and adapt to large angle change caused by large deformation, In this paper, we propose a one-to-one exact matching of the descriptor using MROGH to obtain the displacement vector of the matching point as the driving force. 4) A feature-based Log-Demons fusion algorithm is proposed: the driving force is calculated separately as a constant. In the process of iterative registration, the updating deformation field of Log-Demons is merged in the way of exponential reduction, so that it can give great influence at the beginning of deformation. When the deformation decreases, the registration is completed by the driving force of Log-Demons itself. It not only speeds up the registration process but also reduces the effect of feature point mismatch. In the course of calculating the updated deformation field, it is proposed that the fusion of the driving force of group property and the updated deformation field of Log-Demons itself can keep the differential homeomorphism in Li Qun. Finally, the Log-Demons algorithm model based on driving force is constructed, and the large deformation image registration is realized. The present method is tested on simulated deformation images, real scene large displacement images and brain images, and compared with other methods such as Log-Demonsn SpectralLog-Demonsln LDDMMM. The experimental results show that the proposed method achieves good results. In this paper, not only the large deformation image registration can be completed, but also the calculated deformation field has better local structure and accuracy.
【学位授予单位】：华东师范大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】