Feature-based methods for non-rigid registration frequently encounter the correspondence problem. Regardless of whether points, lines, curves or surface parameterizations are used, feature-based non-rigid matching requires us to automatically solve for correspondences between two sets of features. In addition, there could be many features in either set that have no counterparts in the other. This outlier rejection problem further complicates an already di#cult correspondence problem. We formulate feature-based non-rigid registration as a non-rigid point matching problem. After a careful review of the problem and an in-depth examination of two types of methods previously designed for rigid robust point matching (RPM), we propose a new general framework for non-rigid point matching. We consider it a general framework because it does not depend on any particular form of spatial mapping. We have also developed an algorithm---the TPS-RPM algorithm---with the thin-plate spline (TPS) as the parameterization of the non-rigid spatial mapping and the softassign for the correspondence. The performance of the TPS-RPM algorithm is demonstrated and validated in a series of carefully designed synthetic experiments. In each of these experiments, an empirical comparison with the popular iterated closest point (ICP) algorithm is also provided. Finally, we apply the algorithm to the problem of non-rigid registration of cortical anatomical structures which is required in brain mapping. While these results are somewhat preliminary, they clearly demonstrate the applicability of our approach to real world tasks involving feature-based non-rigid registration.

Abstract—An overview is presented of the medical image processing literature on mutual-information-based registration. The aim of the survey is threefold: an introduction for those new to the field, an overview for those working in the field, and a reference for those searching for literature on a specific application. Methods are classified according to the different aspects of mutual-information-based registration. The main division is in aspects of the methodology and of the application. The part on methodology describes choices made on facets such as preprocessing of images, gray value interpolation, optimization, adaptations to the mutual information measure, and different types of geometrical transformations. The part on applications is a reference of the literature available on different modalities, on interpatient registration and on different anatomical objects. Comparison studies including mutual information are also considered. The paper starts with a description of entropy and mutual information and it closes with a discussion on past achievements and some future challenges. Index Terms—Image registration, literature survey, matching, mutual information. I.

We have evaluated eight different similarity measures used for rigid body registration of serial magnetic resonance (MR) brain scans. To assess their accuracy we used 33 clinical threedimensional (3-D) serial MR images, with deformable extradural tissue excluded by manual segmentation and simulated 3-D MR images with added intensity distortion. For each measure we determined the consistency of registration transformations for both sets of segmented and unsegmented data. We have shown that of the eight measures tested, the ones based on joint entropy produced the best consistency. In particular, these measures seemed to be least sensitive to the presence of extradural tissue. For these data the difference in accuracy of these joint entropy measures, with or without brain segmentation, was within the threshold of visually detectable change in the difference images.

Abstract. Mutual information has been successfully used as an effective similarity measure for multimodal image registration. However, a drawback of the standard mutual information-based computation is that the joint histogram is only calculated from the correspondence between individual voxels in the two images. In this paper, the normalized mutual information measure is extended to consider the correspondence between voxel blocks in multimodal rigid registration. The ambiguity and highdimensionality that appears when dealing with the voxel neighborhood is solved using uniformly distributed random lines and reducing the number of bins of the images. Experimental results show a significant improvement with respect to the standard normalized mutual information. 1

A new algorithm has been developed in this thesis for the non-rigid point matching problem. Designed as an integrated framework, the algorithm jointly estimates a one-to-one correspondence and a non-rigid transformation between two sets of points. The resulting algorithm is called “robust point matching (RPM) algorithm ” because of its capability to tolerate noise and to reject possible outliers existed within the data points. The algorithm is built upon the heuristic of “fuzzy correspondence”, which allows for multiple partial cor-respondences between points. With the help of the deterministic annealing technique, this new heuristic enables the algorithm to overcome many local minima that can be encountered in the matching process. Devised as a general point matching framework, the algorithm can be easily extended to accommodate differ-ent speci£c requirements in many registration applications. Firstly, the modular design of the transformation module enables convenient incorporation of different non-rigid splines. Secondly, the point matching algorithm can be easily extended into a symmetric joint clustering-matching framework. It will be shown that by introducing a super point-set, the joint cluster-matching extension can be applied to estimate an average shape point-set from multiple point shape sets. The algorithm is applied to the registration of 3D brain anatomical structures. We proposed in this work a joint feature registration framework, which is mainly based on the joint clustering-matching extension of the robust

One of the key aspects in 3D-image registration is the computation of the joint intensity histogram. We propose a new approach to compute this histogram using uniformly distributed random lines to sample stochastically the overlapping volume between two 3D-images. The intensity values are captured from the lines at evenly spaced positions, taking an initial random offset different for each line. This method provides us with an accurate, robust and fast mutual information-based registration. The interpolation effects are drastically reduced, due to the stochastic nature of the line generation, and the alignment process is also accelerated. The results obtained show a better performance of the introduced method than the classic computation of the joint histogram.

Two new similarity measures for rigid image registration, based on the normalization of Jensen's difference applied to Renyi and Tsallis-Havrda-Charvat entropies, are introduced. One measure is normalized by the first term of Jensen's difference, which in our proposal coincides with the marginal entropy, and the other by the joint entropy. These measures can be seen as an extension of two measures successfully applied in medical image registration: the mutual information and the normalized mutual information. Experiments with various registration modalities show that the new similarity measures are more robust than the normalized mutual information for some modalities and a determined range of the entropy parameter. Also, a certain improvement on accuracy can be obtained for a different range of this parameter.

this dissertation. 1.1 Radiosity, Complexity, and Information Theory The three fundamental pillars of this thesis are radiosity, complexity, and information theory: One of the most important topics in computer graphics is the accurate computation of the global illumination in a closed virtual environment (scene), i.e. the intensities of light over all its surfaces. "The production of realistic images requires in particular a precise treatment of lighting e#ects that can be achieved by simulating the underlying physical phenomena of light emission, propagation, and reflection"[82]. This type of simulation is called global illumination and is represented by the rendering equation [43], which is a Fredholm integral equation of the second kind. However obtaining an exact representation of the illumination is an intractable problem. Many di#erent techniques are used to obtain an approximate quantification of it [12, 82, 33]

Abstract. Exploded views are often used in illustration to overcome the problem of occlusion when depicting complex structures. In this paper, we propose a volume visualization technique inspired by exploded views that partitions the volume into a number of parallel slabs and shows them apart from each other. The thickness of slabs is driven by the similarity between partitions. We use an information-theoretic technique for the generation of exploded views. First, the algorithm identifies the viewpoint which gives the most structured view of the data. Then, the partition of the volume into the most informative slabs for exploding is obtained using two complementary similarity-based strategies. The number of slabs and the similarity parameter are freely adjustable by the user. 1

Abstract. We present two new clustering algorithms for medical image segmentation based on the multimodal image registration and the information bottleneck method. In these algorithms, the histogram bins of two registered multimodal 3D-images are clustered by minimizing the loss of mutual information between them. Thus, the clustering of histogram bins is driven by the preservation of the shared information between the images, extracting from each image the structures that are more relevant to the other one. In the first algorithm, we segment only one image at a time, while in the second both images are simultaneously segmented. Experiments show the good behavior of the presented algorithms, especially the simultaneous clustering. 1