Simultaneous Registration of Multiple Images: Similarity Metrics and Efficient Optimization

Abstract

We address the alignment of a group of images with simultaneous registration. Therefore, we provide further insights into a recently introduced framework for multivariate similarity measures, referred to as accumulated pair-wise estimates (APE), and derive efficient optimization methods for it. More specifically, we show a strict mathematical deduction of APE from a maximum-likelihood framework and establish a connection to the congealing framework. This is only possible after an extension of the congealing framework with neighborhood information. Moreover, we address the increased computational complexity of simultaneous registration by deriving efficient gradient-based optimization strategies for APE: Gauss-Newton and the efficient second-order minimization (ESM). We present next to SSD the usage of intrinsically nonsquared similarity measures in this least squares optimization framework. The fundamental assumption of ESM, the approximation of the perfectly aligned moving image through the fixed image, limits its application to monomodal registration. We therefore incorporate recently proposed structural representations of images which allow us to perform multimodal registration with ESM. Finally, we evaluate the performance of the optimization strategies with respect to the similarity measures, leading to very good results for ESM. The extension to multimodal registration is in this context very interesting because it offers further possibilities for evaluations, due to publicly available datasets with ground-truth alignment.

Groupwise registration of sequential images from multispectral imaging (MSI) of the retina and choroid

“…It can be used for establishing anatomical correspondences among subjects in a population in order for characterizing anatomical shape differences within/across populations [18,19]. It can also achieve building connections between similar points/regions across different images in order for representation, 3D modeling, synthesis, morphing, browsing of for example human faces [20,21].…”

“…The methods in [25,26] exploit sparse learning and focus on face alignment. The techniques in [19,27] assume a Markov property of the image and accumulate pairwise image matching in order for achieving a groupwise registration.…”

Groupwise non-rigid registration on multiparametric abdominal DWI acquisitions for robust ADC estimation: Comparison with pairwise approaches and different multimodal metrics

“…Most approaches pose the registration problem from a pairwise standpoint [8] using an [ideally] undistorted image as reference; this procedure, however, is prone to an undesired bias towards the a priori chosen template [9], which, depending on its quality, may give rise to multiple outliers in the alignment. On the other hand, groupwise approaches are based on an image reference that is built out of the whole image set to be registered, so that the template bias disappears.…”

Joint alignment of multispectral images via semidefinite programming

“…However, it is not a trivial in practice due to the misalignment between these spectral slices mainly because of the fact that the acquisition time of eye MSI images is commonly longer than the natural time scale of eye's saccadic movement [8]. Similar to the research in other fields [9][10][11][12][13][14][15][16], this spatial misalignment may introduce troubles into not only automatic but also manual interpretation and quantification of the anatomic/metabolic variations between spectral slices.…”

“…There are various group image registration approaches, including the well-known congealing framework and its extensions [20][21][22] which focus on aligning a number of binary images of handwritten digits, 2D face images or 3D medical images via a sequential optimization process to gradually reduce the entropy of image intensity's distribution. Other approaches try to improve alignment performances by taking advantage of shape matching [23], sparse learning [24,25], image's Markov property [14,26], pairwise registration based on normalized cross-correlation [27], robust matching based on gradient descent optimization [28], image contours [29] and joint alignment by enforcing temporal smoothness on motion [30].…”