Progressive refinement for robust image registration

A new image registration algorithm with robust cost function and progressive refinement estimation is developed on the basis of direct method (DM). The robustness lies in M-estimation to avert larger local noise and outliers.     

Efficient subpixel image registration algorithms

Three new algorithms for 2D translation image registration to within a small fraction of a pixel that use nonlinear optimization and matrix-multiply discrete Fourier transforms are compared. These algorithms can achieve registration with an accuracy equivalent to that of the conventional fast Fourier transform upsampling approach in a small fraction of the computation time and with greatly reduced memory requirements. Their accuracy and computation time are compared for the purpose of evaluating a translation-invariant error metric.     

远距离多源图像融合系统实时配准设计

为了实现全天候多波段远距离实时图像监控,设计了具有微光、红外和可见光融合的光学前端,对多源图像进行实时配准研究。在平行光轴的基础上,通过计算不同视场图像的成像视差,计算仿射变换需要的参数,采用双线性内插算法弥补红外在成像方面与可见光图像的差别,将红外图像的变换制作成查找表存储在图像处理器DM642中,系统通过硬件查找表可以快速实现不同图像的配准变换,实现同步视频的实时配准与融合。实验表明:该设计能够准确地实现多源图像的实时配准,系统经过图像配准、图像融合和伪彩变换处理后的时间约为24.3 ms,系统探测距离大于3 km。     

Tomotherapy planning and image registration for planning CT image slice thickness for stereotactic radiosurgery

Journal of the Korean Physical Society - Tomotherapy (TomoTherapy, Inc., Madison, WI, USA) is a helical intensity modulated radiation therapy (IMRT) modality using rotational delivery of a fan beam...     

针对图像序列的快速低存储开销配准算法

针对序列图像配准问题，提出一种快速低存储开销配准算法。首先，生成一系列与具体图像内容无关的特征点。而后，使用正向-反向跟踪来获取稳定的特征点对,其中，正向跟踪用于获得所有可能的特征点对，反向跟踪用来得到正向-反向误差，并且利用此误差来获取最终稳定的特征点对。最后，在稳定特征点对的基础上通过归一化直接线性变换计算得到可用于图像配准的单应矩阵。实验表明该算法能够提供与优秀的传统算法相当的配准性能。由于该算法对序列中图像之间的连续性进行了充分利用，不仅降低了存储开销，还提高了运算速度。对480360的序列而言，该算法需要的存储开销仅为421 kB，且运算速度达到32 帧/s。     

基于特征点的对数极坐标变换图像配准算法

介绍了一种新的图像配准算法,可以很好地解决图像配准中的平移、旋转和缩放问题。算法的实现是首先使用SUSAN算子提取两幅图像的特征角点,剔除虚假的特征点(噪声点),然后使用改进对数极坐标变换和投影相关匹配算法实现特征点匹配;结合亚像素定位技术,可以进一步提高算法的精度。     

针对图像序列的快速低存储开销配准算法

针对序列图像配准问题,提出一种快速低存储开销配准算法。首先,生成一系列与具体图像内容无关的特征点。而后,使用正向-反向跟踪来获取稳定的特征点对,其中,正向跟踪用于获得所有可能的特征点对,反向跟踪用来得到正向-反向误差,并且利用此误差来获取最终稳定的特征点对。最后,在稳定特征点对的基础上通过归一化直接线性变换计算得到可用于图像配准的单应矩阵。实验表明该算法能够提供与优秀的传统算法相当的配准性能。由于该算法对序列中图像之间的连续性进行了充分利用,不仅降低了存储开销,还提高了运算速度。对480360的序列而言,该算法需要的存储开销仅为421 kB,且运算速度达到32 帧/s。     

Subpixel image registration with reduced bias

A backward linear digital image correlation algorithm was introduced to obtain subpixel image registration without noise-induced bias for an image set consisting of a noise-free reference image and a number of noisy current images. Furthermore, a correction procedure using additional reference images (generated by offsetting the original image to displacement increments of either half-pixels or even quarter-pixels) was proposed to reduce subpixel approximation bias in the analysis. Numerical results of six sets of synthetic images showed that the proposed algorithm was effective in improving the accuracy of subpixel image registration.     

改进的SURF彩色遥感图像配准算法

为了进一步提高彩色遥感图像的配准精度,针对遥感图像配准过程中色彩信息利用率低以及误匹配率高的问题,提出一种改进的SURF(Speeded Up Robust Feature)彩色遥感图像配准算法。该算法首先在对彩色遥感图像进行特征点检测基础上,对特征点描述算子进行改进,以使颜色空间变换后得到的特征点色彩信息添加到原描述算子中,并对特征点描述算子进行归一化处理,以增加算子的独特性和对旋转、尺度、光照的鲁棒性;其次,结合单向匹配和双向匹配的比值,提出了一种最优化阈值选择准则,如果欧氏距离比率小于最优化阈值,完成特征点匹配,得到正确匹配点对,再通过变换矩阵得到配准图像。实验结果表明,在保证实时性的条件下,该算法相比于原SURF算法,准确性和稳定性都有一定提高,具有一定的理论和应用价值。     

图像配准中几种特征点提取方法的分析与实验

研究了图像配准中常用的几种特征点提取方法。基于图像配准对特征提取方法的特殊要求,提出了一组量化的特征提取方法评价因子,能用于合理评价各种方法的性能、优点与缺点,为这些方法的改进提供指导性意见和比较依据。通过大量的实验,利用评价因子比较和分析了上述的几种常用特征提取方法的优缺点、适用环境,并提出改进建议。     

新的Cauchy-Schwarz距离函数与多模态医学图像配准

信息论测度,特别是Shannon互信息是多模态图像配准的一种重要方法,但除了互信息之外,仍然存在其它的函数来实现这一任务。对互信息、Kullback-Leibler距离和Shannon不等式之间相互关系作了分析,根据这些关系和不等式理论,提出了新的Cauchy-Schwarz距离,并将这一距离测度用于多模态医学图像的配准处理。从计算速度、噪声容忍性、测度函数图形的特点和图像窗口大小影响等几个方面,通过MR和PET医学图像的实验分析,对新的Cauchy-Schwarz距离测度和典型的Shannon信息论测度进行了分析比较。实验结果表明,新的Cauchy-Schwarz距离测度函数同样可以用于多模态图像配准,而且有着更强的噪声容忍性和更为节省的计算量。     

Application of three-dimensional magnetic resonance image registration for monitoring hip joint diseases

The purpose of this study was to estimate the accuracy of a method in which three-dimensional (3D) magnetic resonance (MR) volume registration is used for monitoring hip joint disease. Data were analyzed using a normalized cross-correlation (NCC) algorithm involving a user-selected 3D box including the proximal femur. Most of the femoral head was not included in the 3D box because it can become deformed during the course of disease. The accuracy of registration around the femoral head was evaluated using five phantoms and clinical MR data of 17 patients with hip joint disease. In the phantom experiment, registration accuracy was evaluated using four fiducial markers attached to the femoral head. In the experiment using clinical data, registration accuracy was evaluated using a landmark in the femoral head. The registration accuracy in the phantom and clinical experiment was 0.43+/-0.18 mm (S.D.) and 1.12+/-0.46 mm (S.D.), respectively. The former is a value less than half the minimum dimension of a voxel (1.25 x 1.25 x 1.0 mm). Although the latter is slightly larger than the minimum dimension of a voxel, actual errors would be smaller because of the uncertainty in landmark localization. In conclusion, the present method based on an NCC algorithm can be used to accurately register serial MR images of the femoral heads with an error on the order of a voxel. We believe that this method is sufficiently accurate for monitoring hip joint diseases.     

基于算术-几何均值距离的多模态图像配准

根据图像灰度联合概率分布函数与图像相似程度之间的关系,提出了一种基于算术 几何均值距离的多模态图像配准新测度。与基于信息论的测度不同,新测度不再要求概率分布必须满足连续性的要求。实验结果表明,所提出的新测度比基于信息论的测度具有更强的噪声鲁棒性和计算量更小。     

一种基于改进频域图像配准技术的超分辨力图像处理方法

为提高光电成像系统的空间分辨力,提出了一种基于改进的频率域图像配准技术的超分辨力图像处理方法。首先利用改进的频域图像配准方法估算出低分辨力图像之间的微位移量,然后采用Papoulis-Gerchberg超分辨力处理方法完成图像复原。利用不同重构方法进行了仿真及实验研究,给出了评价参数。模拟和实际显微热图像的处理结果表明:该算法可使图像质量得到改善,分辨的细节更多,可有效地提高光电成像系统的空间分辨力;处理算法简单,计算量小,可实现快速处理。该算法还可应用于其他不可控光学微扫描成像系统中,具有广泛的应用前景。     

基于小波变换的脑部医学Demons图像配准

非刚性配准是医学图像处理的一个重要研究方向。针对Demons衍生出的一系列经典的配准算法在医学图像应用上计算复杂、方向信息不足问题进行了研究。基于光流场模型的Demons算法依赖图像灰度梯度是图像发生变形,当缺乏梯度信息时,力不能确定,因而容易造成误差,并且该算法仅适合于单模态图像配准。为此本文提出了一种基于小波变换理论的频域Demons配准处理方法(B-Demons)。该方法利用小波变换能够对各个尺度、方向和位置实现较好定位的优势,通过高频、低频的图像变换反映出图像的特征信息。实验结果证明了算法的有效性和鲁棒性。     

3D image registration using a fast noniterative algorithm

This note describes the implementation of a three-dimensional (3D) registration algorithm, generalizing a previous 2D version [Alexander, Int J Imaging Systems and Technology 1999;10:242-57]. The algorithm solves an integrated form of linearized image matching equation over a set of 3D rectangular sub-volumes ('patches') in the image domain. This integrated form avoids numerical instabilities due to differentiation of a noisy image over a lattice, and in addition renders the algorithm robustness to noise. Registration is implemented by first convolving the unregistered images with a set of computationally fast [O(N)] filters, providing four bandpass images for each input image, and integrating the image matching equation over the given patch. Each filter and each patch together provide an independent set of constraints on the displacement field derived by solving a set of linear regression equations. Furthermore, the filters are implemented at a variety of spatial scales, enabling registration parameters at one scale to be used as an input approximation for deriving refined values of those parameters at a finer scale of resolution. This hierarchical procedure is necessary to avoid false matches occurring. Both downsampled and oversampled (undecimating) filtering is implemented. Although the former is computationally fast, it lacks the translation invariance of the latter. Oversampling is required for accurate interpolation that is used in intermediate stages of the algorithm to reconstruct the partially registered from the unregistered image. However, downsampling is useful, and computationally efficient, for preliminary stages of registration when large mismatches are present. The 3D registration algorithm was implemented using a 12-parameter affine model for the displacement: u(x) = Ax + b. Linear interpolation was used throughout. Accuracy and timing results for registering various multislice images, obtained by scanning a melon and human volunteers in various stationary positions, is described. The algorithm may be generalized to more general models of the displacement field, and is also well suited to parallel processing.     

采用整体最小二乘法的条纹图配准方法

针对共光路菲索型动态干涉仪采集到的4幅条纹图的空间一致性问题,本文提出一种新的条纹图配准思路,将条纹图与图像配准在关联度上人为分开,搭建了专用的图像配准装置,避免将条纹与十字丝标志点混在一幅图像内而影响干涉仪的测量精度。首先通过本文搭建的条纹图配准装置对4台CMOS相机进行物理配准,然后利用整体最小二乘法对采集到的同一十字丝刻划板图像进行十字丝提取、交点计算以及旋转量计算,实现共光路菲索型动态干涉仪条纹图像的点点对应。最后通过试验对比验证,证明了本文算法的配准精度优于模板重心法的配准结果,互相关度达到96%以上。     

An improvement of image registration based on phase correlation

This paper is dedicated to studying the extended phase correlation method that applies to the sub-pixel evaluation of translational displacement between images. On the basis of experiments, this paper proposes an improved version of classical Hanning window, with involving parameters optimization in the process. We add a window function to the input images in the space domain and a weighting function to the spectrum in the frequency domain. The statistics and distribution of errors prove the applicability of the proposal. The experimental results of the proposed extended phase correlation algorithm show the registration accuracy of 1/100 pixel.     

Adaptive optics retinal image registration from scale-invariant feature transform

With the use of adaptive optics(AO), high-resolution microscopic imaging of the living human retina in the single cell level has been achieved. In an AO retinal imaging system, with a small field size (about 1°, 300 μm) the motion of the eye severely affects the stabilization of the real-time video images results in significant distortions of the retina images. Scale-invariant feature transform (SIFT) algorithm is applied to automatically abstract corner points with subpixel resolution and match the points in two frames. With the matched corner points, we estimate and remove the motions of 20 frames of photoreceptor cells and capillary blood vessels, respectively. The maximal translational motion is about 30 and 44 pixels in the 20 frames whose size is 416 × 416 pixels. More general motions can be considered by the SIFT algorithm, but only simple translational motion can be considered by cross-correlation algorithm.