-norm) is unable to correctly represent non-smooth displacement fields, that can, for example, occur at sliding interfaces in the thorax and abdomen in image time-series during respiration. In this paper, isotropic Total Variation (TV) regularization is used to enable accurate registration near such interfaces. We further develop the TV-regularization for parametric displacement fields and provide an efficient numerical solution scheme using the Alternating Directions Method of Multipliers (ADMM). The proposed method was successfully applied to four clinical databases which capture breathing motion, including CT lung and MR liver images. It provided accurate registration results for the whole volume. A key strength of our proposed method is that it does not depend on organ masks that are conventionally required by many algorithms to avoid errors at sliding interfaces. Furthermore, our method is robust to parameter selection, allowing the use of the same parameters for all tested databases. The average target registration error (TRE) of our method is superior (10% to 40%) to other techniques in the literature. It provides precise motion quantification and sliding detection with sub-pixel accuracy on the publicly available breathing motion databases (mean TREs of 0.95 mm for DIR 4D CT, 0.96 mm for DIR COPDgene, 0.91 mm for POPI databases).

Variations in the shape and appearance of anatomical structures in medical images are often relevant radiological signs of disease. Automatic tools can help automate parts of this manual process. A cloud-based evaluation framework is presented in this paper including results of benchmarking current state-of-the-art medical imaging algorithms for anatomical structure segmentation and landmark detection: the VISCERAL Anatomy benchmarks. The algorithms are implemented in virtual machines in the cloud where participants can only access the training data and can be run privately by the benchmark administrators to objectively compare their performance in an unseen common test set. Overall, 120 computed tomography and magnetic resonance patient volumes were manually annotated to create a standard Gold Corpus containing a total of 1295 structures and 1760 landmarks. Ten participants contributed with automatic algorithms for the organ segmentation task, and three for the landmark localization task. Different algorithms obtained the best scores in the four available imaging modalities and for subsets of anatomical structures. The annotation framework, resulting data set, evaluation setup, results and performance analysis from the three VISCERAL Anatomy benchmarks are presented in this article. Both the VISCERAL data set and Silver Corpus generated with the fusion of the participant algorithms on a larger set of non-manually-annotated medical images are available to the research community.

In this paper, a novel Markov random field (MRF)-based approach is presented for segmenting medical images while simultaneously registering an atlas nonrigidly. In the literature, both segmentation and registration have been studied extensively. For applications that involve both, such as segmentation via atlas-based registration, earlier studies proposed addressing these problems iteratively by feeding the output of each to initialize the other. This scheme, however, cannot guarantee an optimal solution for the combined task at hand, since these two individual problems are then treated separately. In this paper, we formulate simultaneous registration and segmentation (SRS) as a maximum a-posteriori (MAP) problem. We decompose the resulting probabilities such that the MAP inference can be done using MRFs. An efficient hierarchical implementation is employed, allowing coarse-to-fine registration while estimating segmentation at pixel level. The method is evaluated on two clinical data sets: 1) mandibular bone segmentation in 3D CT and 2) corpus callosum segmentation in 2D midsaggital slices of brain MRI. A video tracking example is also given. Our implementation allows us to directly compare the proposed method with the individual segmentation/registration and the iterative approach using the exact same potential functions. In a leave-one-out evaluation, SRS demonstrated more accurate results in terms of dice overlap and surface distance metrics for both data sets. We also show quantitatively that the SRS method is less sensitive to the errors in the registration as opposed to the iterative approach.

We present latent log-linear models, an extension of log-linear models incorporating latent variables, and we propose two applications thereof: log-linear mixture models and image deformation-aware log-linear models. The resulting models are fully discriminative, can be trained efficiently, and the model complexity can be controlled. Log-linear mixture models offer additional flexibility within the log-linear modeling framework. Unlike previous approaches, the image deformation-aware model directly considers image deformations and allows for a discriminative training of the deformation parameters. Both are trained using alternating optimization. For certain variants, convergence to a stationary point is guaranteed and, in practice, even variants without this guarantee converge and find models that perform well. We tune the methods on the USPS data set and evaluate on the MNIST data set, demonstrating the generalization capabilities of our proposed models. Our models, although using significantly fewer parameters, are able to obtain competitive results with models proposed in the literature.

In this work, we present novel warping algorithms for full 2D pixel-grid deformations for face recognition. Due to high variation in face appearance, face recognition is considered a very difficult task, especially if only a single reference image, for example a mug-shot, per face is available. Usually model-based approaches with additional training data are used to cope with several types of variation occurring in facial imaging. Image warping contrarily yields a distance measure which is invariant with regard to several types of variation. This allows for precise recognition even using only very few reference observations. Due to the computationally complex problem of optimal 2D warping, pseudo-2D warping-based approaches in the past represented strong approximations of the original problem, and were mainly successful on data with low variability or rectified images. We propose a novel 2D warping method which is globally optimal and makes no prior assumtions on the data variability besides two-dimensional smootheness constraints which both avoid local mirroring and gaps and significantly speed up the optimization. Furthermore, we show that occlusion handling is imperative to obtain smooth warpings in a variety of domains. We evaluate our novel algorithm on various well known databases, such as the AR-Face and CMU-PIE database, and provide a detailed comparison to existing warping approaches. We show that by using simple relative 2D constraints, strong local features and a kernel, which is robust w.r.t. occlusions, our computationally complex approaches outperform state-of-the-art results for recognizing faces under varying expressions, occlusions and poses. Most interestingly, we achieve higher accuracy using fewer training instances per class compared to methods learning a model of the 3D shape.

Conventional speech recognition systems are based on hidden Markov models (HMMs) with Gaussian mixture models (GHMMs). Discriminative log-linear models are an alternative modeling approach and have been investigated recently in speech recognition. GHMMs are directed models with constraints, e.g., positivity of variances and normalization of conditional probabilities, while log-linear models do not use such constraints. This paper compares the posterior form of typical generative models related to speech recognition with their log-linear model counterparts. The key result will be the derivation of the equivalence of these two different approaches under weak assumptions. In particular, we study Gaussian mixture models, part-of-speech bigram tagging models, and eventually, the GHMMs. This result unifies two important but fundamentally different modeling paradigms in speech recognition on the functional level. Furthermore, this paper will present comparative experimental results for various speech tasks of different complexity, including a digit string and large-vocabulary continuous speech recognition tasks.

We propose to use energy minimization in MRFs for matching-based image recognition tasks. To this end, the Tree-Reweighted Message Passing algorithm is modified by geometric constraints and efficiently used by exploiting the guaranteed monotonicity of the lower bound within a nearest-neighbor based classification framework. The constraints allow for a speedup linear to the dimensionality of the reference image, and the lower bound allows to optimally prune the nearest-neighbor search without loosing accuracy, effectively allowing to increase the number of optimization iterations without an effect on runtime. We evaluate our approach on well-known OCR and face recognition tasks and on the latter outperform current state-of-the-art.