Large-scale datasets are foundational for research and development in natural language processing. However, current approaches face three key challenges: (1) reliance on ambiguously licensed sources restricting use, sharing, and derivative works; (2) static dataset releases that prevent community contributions and diminish longevity; and (3) quality assurance processes restricted to publishing teams rather than leveraging community expertise. To address these limitations, we introduce two contributions: the Dynaword approach and Danish Dynaword. The Dynaword approach is a framework for creating large-scale, open datasets that can be continuously updated through community collaboration. Danish Dynaword is a concrete implementation that validates this approach and demonstrates its potential. Danish Dynaword contains over four times as many tokens as comparable releases, is exclusively openly licensed, and has received multiple contributions across industry and research. The repository includes light-weight tests to ensure data formatting, quality, and documentation, establishing a sustainable framework for ongoing community contributions and dataset evolution.

We consider the problem of acquiring causal representations and concepts in a reinforcement learning setting. Our approach defines a causal variable as being both manipulable by a policy, and able to predict the outcome. We thereby obtain a parsimonious causal graph in which interventions occur at the level of policies. The approach avoids defining a generative model of the data, prior pre-processing, or learning the transition kernel of the Markov decision process. Instead, causal variables and policies are determined by maximizing a new optimization target inspired by mediation analysis, which differs from the expected return. The maximization is accomplished using a generalization of Bellman's equation which is shown to converge, and the method finds meaningful causal representations in a simulated environment.

We present a new method for automatic classification of indirect immunoflourescence images of HEp-2 cells into different staining pattern classes. Our method is based on a new texture measure called shape index histograms that captures second-order image structure at multiple scales. Moreover, we introduce a spatial decomposition scheme which is radially symmetric and suitable for cell images. The spatial decomposition is performed using donut-shaped pooling regions of varying sizes when gathering histogram contributions. We evaluate our method using both the ICIP 2013 and the ICPR 2012 competition datasets. Our results show that shape index histograms are superior to other popular texture descriptors for HEp-2 cell classification. Moreover, when comparing to other automated systems for HEp-2 cell classification we show that shape index histograms are very competitive; especially considering the relatively low complexity of the method.

This paper deals with the problem of generating quality tetrahedral meshes for image registration. From an initial coarse mesh the approach matches the mesh to the image volume by combining red-green subdivision and mesh evolution through mesh-to-image matching regularized with a mesh quality measure. The method was tested on a T1 weighted MR volume of an adult brain and showed a 66% reduction in the number of mesh vertices compared to a red-subdivision strategy. The deformation capability of the mesh was tested by registration to five additional T1-weighted MR volumes.

A custom designed markerless tracking system was demonstrated to be applicable for positron emission tomography (PET) brain imaging. Precise head motion registration is crucial for accurate motion correction (MC) in PET imaging. State-of-the-art tracking systems applied with PET brain imaging rely on markers attached to the patient's head. The marker attachment is the main weakness of these systems. A healthy volunteer participating in a cigarette smoking study to image dopamine release was scanned twice for 2 h with (11)C-racolopride on the high resolution research tomograph (HRRT) PET scanner. Head motion was independently measured, with a commercial marker-based device and the proposed vision-based system. A list-mode event-by-event reconstruction algorithm using the detected motion was applied. A phantom study with hand-controlled continuous random motion was obtained. Motion was time-varying with long drift motions of up to 18 mm and regular step-wise motion of 1-6 mm. The evaluated measures were significantly better for motion-corrected images compared to no MC. The demonstrated system agreed with a commercial integrated system. Motion-corrected images were improved in contrast recovery of small structures.

We present a system for head motion tracking in 3D brain imaging. The system is based on facial surface reconstruction and tracking using a structured light (SL) scanning principle. The system is designed to fit into narrow 3D medical scanner geometries limiting the field of view. It is tested in a clinical setting on the high resolution research tomograph (HRRT), Siemens PET scanner with a head phantom and volunteers. The SL system is compared to a commercial optical tracking system, the Polaris Vicra system, from NDI based on translatory and rotary ground truth motions of the head phantom. The accuracy of the systems was similar, with root mean square (rms) errors of 0.09 degrees for ±20 degrees axial rotations, and rms errors of 0.24 mm for ± 25 mm translations. Tests were made using (1) a light emitting diode (LED) based miniaturized video projector, the Pico projector from Texas Instruments, and (2) a customized version of this projector replacing a visible light LED with a 850 nm near infrared LED. The latter system does not provide additional discomfort by visible light projection into the patient's eyes. The main advantage over existing head motion tracking devices, including the Polaris Vicra system, is that it is not necessary to place markers on the patient. This provides a simpler workflow and eliminates uncertainties related to marker attachment and stability. We show proof of concept of a marker less tracking system especially designed for clinical use with promising results.

Remarkable results have been obtained using image models based on image patches, for example sparse generative models for image inpainting, noise reduction and superresolution, sparse texture segmentation or texton models.In this paper we propose a powerful and yet simple approach for segmentation using dictionaries of image patches with associated label data.The approach is based on ideas from sparse generative image models and texton based texture modeling.The intensity and label dictionaries are learned from training images with associated label information of (a subset) of the pixels based on a modified vector quantization approach.For new images the intensity dictionary is used to encode the image data and the label dictionary is used to build a segmentation of the image.We demonstrate the algorithm on composite and real texture images and show how successful training is possible even for noisy image and low-quality label training data.In our experimental evaluation we achieve state-of-the-art performance for segmentation.

This paper presents a fusion of the active appearance model (AAM) and the Riemannian elasticity framework which yields a non-linear shape model and a linear texture model -the active elastic appearance model (EAM).The non-linear elasticity shape model is more flexible than the usual linear subspace model, and it is therefore able to capture more complex shape variations.Local rotation and translation invariance are the primary explanation for the additional flexibility.In addition, we introduce global scale invariance into the Riemannian elasticity framework which together with the local translation and rotation invariances eliminate the need for separate pose estimation.The new approach was tested against AAM in three experiments; face labeling, face labeling with poor initialization and corpus callosum segmentation.In all the examples the EAM performed significantly better than AAM.Our Matlab implementation can be downloaded through svn from https://svn.imm.dtu.dk/AAMLab/svn/AAMLab/trunk/ .

Abstract A growing number of applications depend on accurate and fast 3D scene analysis. Examples are model and lightfield acquisition, collision prevention, mixed reality and gesture recognition. The estimation of a range map by image analysis or laser scan techniques is still a time‐consuming and expensive part of such systems. A lower‐priced, fast and robust alternative for distance measurements are time‐of‐flight (ToF) cameras. Recently, significant advances have been made in producing low‐cost and compact ToF devices, which have the potential to revolutionize many fields of research, including computer graphics, computer vision and human machine interaction (HMI). These technologies are starting to have an impact on research and commercial applications. The upcoming generation of ToF sensors, however, will be even more powerful and will have the potential to become ‘ubiquitous real‐time geometry devices’ for gaming, web‐conferencing, and numerous other applications. This paper gives an account of recent developments in ToF technology and discusses the current state of the integration of this technology into various graphics‐related applications.

A method for implicit surface reconstruction is proposed. The novelty in this paper is the adaptation of Markov Random Field regularization of a distance field. The Markov Random Field formulation allows us to integrate both knowledge about the type of surface we wish to reconstruct (the prior) and knowledge about data (the observation model) in an orthogonal fashion. Local models that account for both scene-specific knowledge and physical properties of the scanning device are described. Furthermore, how the optimal distance field can be computed is demonstrated using conjugate gradients, sparse Cholesky factorization, and a multiscale iterative optimization scheme. The method is demonstrated on a set of scanned human heads and, both in terms of accuracy and the ability to close holes, the proposed method is shown to have similar or superior performance when compared to current state-of-the-art algorithms.

Many change detection algorithms work by calculating the probability of change on a pixel-wise basis. This is a disadvantage since one is usually looking for regions of change, and such information is not used in pixel-wise classification - per definition. This issue becomes apparent in the face of noise, implying that the pixel-wise classifier is also noisy. There is thus a need for incorporating local homogeneity constraints into such a change detection framework. For this modelling task Markov Random Fields are suitable. Markov Random Fields have, however, previously been plagued by lack of efficient optimization methods or numerical solvers. We here address the issue of efficient incorporation of local homogeneity constraints into change detection algorithms. We do this by exploiting recent advances in graph based algorithms for Markov Random Fields. This is combined with an IR-MAD change detector, and demonstrated on real data with good results.

Abstract Scalar functions defined on manifold triangle meshes is a starting point for many geometry processing algorithms such as mesh parametrization, skeletonization, and segmentation. In this paper, we propose the Auto Diffusion Function (ADF) which is a linear combination of the eigenfunctions of the Laplace‐Beltrami operator in a way that has a simple physical interpretation. The ADF of a given 3D object has a number of further desirable properties: Its extrema are generally at the tips of features of a given object, its gradients and level sets follow or encircle features, respectively, it is controlled by a single parameter which can be interpreted as feature scale, and, finally, the ADF is invariant to rigid and isometric deformations. We describe the ADF and its properties in detail and compare it to other choices of scalar functions on manifolds. As an example of an application, we present a pose invariant, hierarchical skeletonization and segmentation algorithm which makes direct use of the ADF.

We present an adaptive parametrization scheme for dynamic mesh refinement in the application of parametric image registration. The scheme is based on a refinement measure ensuring that the control points give an efficient representation of the warp fields, in terms of minimizing the registration cost function. In the current work we introduce multivariate B-splines as a novel alternative to the widely used tensor B-splines enabling us to make efficient use of the derived measure.The multivariate B-splines of order n are Cn-1smooth and are based on Delaunay configurations of arbitrary 2D or 3D control point sets. Efficient algorithms for finding the configurations are presented, and B-splines are through their flexibility shown to feature several advantages over the tensor B-splines. In spite of efforts to make the tensor product B-splines more flexible, the knots are still bound to reside on a regular grid. In contrast, by efficient non- constrained placement of the knots, the multivariate B- splines are shown to give a good representation of inho- mogeneous objects in natural settings. The wide applicability of the method is illustrated through its application on medical data and for optical flow estimation.

Nonlinear registration is mostly performed after initialization by a global, linear transformation (in this work, we focus on similarity transformations), computed by a linear registration method. For the further processing of the results, it is mostly assumed that this preregistration step completely removes the respective linear transformation. However, we show that in deformable settings, this is not the case. As a consequence, a significant linear component is still existent in the deformation computed by the nonlinear registration algorithm. For construction of statistical shape models (SSM) from deformations, this is an unwanted property: SSMs should not contain similarity transformations, since these do not capture information about shape. We propose a method which performs an a posteriori extraction of a similarity transformation from a given nonlinear deformation field, and we use the processed fields as input for SSM construction. For computation of minimal displacements, a closed-form solution minimizing the squared Euclidean norm of the displacement field subject to similarity parameters is used. Experiments on real inter-subject data and on a synthetic example show that the theoretically justified removal of the similarity component by the proposed method has a large influence on the shape model and significantly improves the results.

Face recognition systems are typically required to work under highly varying illumination conditions. This leads to complex effects imposed on the acquired face image that pertains little to the actual identity. Consequently, illumination normalization is required to reach acceptable recognition rates in face recognition systems. In this paper, we propose an approach that integrates the face identity and illumination models under the widely used active appearance model framework as an extension to the texture model in order to obtain illumination-invariant face localization.

In this paper we present a new method for constructing diffeomorphic statistical deformation models in arbitrary dimensional images with a nonlinear generative model and a linear parameter space. Our deformation model is a modified version of the diffeomorphic model introduced by Cootes et al. The modifications ensure that no boundary restriction has to be enforced on the parameter space to prevent folds or tears in the deformation field. For straightforward statistical analysis, principal component analysis and sparse methods, we assume that the parameters for a class of deformations lie on a linear manifold and that the distance between two deformations are given by the metric introduced by the L2-norm in the parameter space. The chosen L2-norm is shown to have a clear and intuitive interpretation on the usual nonlinear manifold. Our model is validated on a set of MR images of corpus callosum with ground truth in form of manual expert annotations, and compared to Cootes's model. We anticipate applications in unconstrained diffeomorphic synthesis of images, e.g. for tracking, segmentation, registration or classification purposes.

Recent advances in statistics have spawned powerful methods for regression and data decomposition that promote sparsity, a property that facilitates interpretation of the results. Sparse models use a small subset of the available variables and may perform as well or better than their full counterparts if constructed carefully. In most medical applications, models are required to have both good statistical performance and a relevant clinical interpretation to be of value. Morphometry of the corpus callosum is one illustrative example. This paper presents a method for relating spatial features to clinical outcome data. A set of parsimonious variables is extracted using sparse principal component analysis, producing simple yet characteristic features. The relation of these variables with clinical data is then established using a regression model. The result may be visualized as patterns of anatomical variation related to clinical outcome. In the present application, landmark-based shape data of the corpus callosum is analyzed in relation to age, gender, and clinical tests of walking speed and verbal fluency. To put the data-driven sparse principal component method into perspective, we consider two alternative techniques, one where features are derived using a model-based wavelet approach, and one where the original variables are regressed directly on the outcome.

Combined modeling of pixel intensities and shape has proven to be a very robust and widely applicable approach to interpret images. As such the active appearance model (AAM) framework has been applied to a wide variety of problems within medical image analysis. This paper summarizes AAM applications within medicine and describes a public domain implementation, namely the flexible appearance modeling environment (FAME). We give guidelines for the use of this research platform, and show that the optimization techniques used renders it applicable to interactive medical applications. To increase performance and make models generalize better, we apply parallel analysis to obtain automatic and objective model truncation. Further, two different AAM training methods are compared along with a reference case study carried out on cross-sectional short-axis cardiac magnetic resonance images and face images. Source code and annotated data sets needed to reproduce the results are put in the public domain for further investigation.

We extend a series of multivariate Bayesian two-dimensional (2-D) contextual classifiers to three-dimensional (3-D) by specifying a simultaneous Gaussian distribution for the feature vectors as well as a prior distribution of the class variables of a pixel and its six nearest 3-D neighbors.

The estimation of flow fields from time sequences of satellite imagery has a number of important applications. For visualization of cloud or sea ice movements in sequences of crude temporal sampling, a satisfactory nonblurred temporal interpolation can be performed only when the flow field or an estimate thereof is known. Estimated flow fields in weather satellite imagery might also be used on an operational basis as inputs to short-term weather prediction. The authors describe a method for the estimation of dense flow fields. Local measurements of motion are obtained by analysis of the local energy distribution, which is sampled by using a set of three-dimensional (3D) spatio-temporal filters. The estimated local energy distribution also allows the authors to compute a confidence measure of the estimated local normal flow. The algorithm, furthermore, utilizes Markovian random fields in order to integrate the local estimates of normal flows into a dense flow field by using measures of spatial smoothness. To obtain smoothness, the authors will constrain first-order derivatives of the flow field. The performance of the algorithm is illustrated by the estimation of the flow fields corresponding to a sequence of Meteosat thermal images. The estimated flow fields are used in a temporal interpolation scheme.