VMV: Vision, Modeling, and Visualization

Permanent URI for this community

https://diglib.eg.org/handle/10.2312/1013

Browse

Now showing 1 - 20 of 403

Coherent Background Video Inpainting through Kalman Smoothing along Trajectories
(The Eurographics Association, 2010) Bugeau, Aurélie; Gargallo, Paul; D'Hondt, Olivier; Hervieu, Alexandre; Papadakis, Nicolas; Caselles, Vicent; Reinhard Koch and Andreas Kolb and Christof Rezk-Salama
Video inpainting consists in recovering the missing or corrupted parts of an image sequence so that the reconstructed sequence looks natural. For each frame, the reconstruction has to be spatially coherent with the rest of the image and temporally with respect to the reconstructions of adjacent frames. Most of existing methods only focus on inpainting foreground objects moving with a periodic motion and consider that the background is almost static. In this paper we address the problem of background inpainting and propose a method that handles dynamic background (illumination changes, moving camera, dynamic textures...). The algorithm starts by applying an image inpainting technique to each frame of the sequence and then temporally smoothes these reconstructions through Kalman smoothing along the estimated trajectories of the unknown points. The computation of the trajectories relies on the estimation of forward and backward dense optical flow fields. Several experiments and comparisons demonstrate the performance of the proposed approach.
Gerbil - A Novel Software Framework for Visualization and Analysis in the Multispectral Domain
(The Eurographics Association, 2010) Jordan, Johannes; Angelopoulou, Elli; Reinhard Koch and Andreas Kolb and Christof Rezk-Salama
Multispectral imaging has been gaining popularity and has been gradually applied to many fields besides remote sensing. Multispectral data provides unique information about material classification and reflectance analysis in general. However, due to the high dimensionality of the data, both human observers as well as computers, have difficulty interpreting this wealth of information. We present a new software package that facilitates the visualization of the relationship between spectral and topological information in a novel fashion. It puts emphasis on the spectral gradient, which is shown to provide enhanced information for many reflectance analysis tasks. It also includes a rich toolbox for evaluation of image segmentation and other algorithms in the multispectral domain. We combine the parallel coordinates visualization technique with hashing for a highly interactive visual connection between spectral distribution, spectral gradient and topology. The framework is released as open-source, has a modern cross-platform design and is well integrated into existing established computer vision software (OpenCV).
Dynamic Focus + Context for Volume Rendering
(The Eurographics Association, 2010) Sikachev, Peter; Rautek, Peter; Bruckner, Stefan; Gröller, M. Eduard; Reinhard Koch and Andreas Kolb and Christof Rezk-Salama
Interactive visualization is widely used in many applications for efficient representation of complex data. Many techniques make use of the focus+context approach in a static manner. These techniques do not fully make use of the interaction semantics. In this paper we present a dynamic focus+context approach that highlights salient features during user interaction. We explore rotation, panning, and zooming interaction semantics and propose several methods of changing visual representations, based on a suggested engagement-estimation method. We use DVR-MIP interpolation and a radial opacity-change approach, exploring rotation, panning, and zooming semantics. Our approach adds short animations during user interaction that help to explore the data efficiently and aid the user in the detection of unknown features.
Real-time Full-body Visual Traits Recognition from Image Sequences
(The Eurographics Association, 2010) Jung, Christoph; Tausch, Reimar; Wojek, Christian; Reinhard Koch and Andreas Kolb and Christof Rezk-Salama
The automatic recognition of human visual traits from images is a challenging computer vision task. Visual traits describe for example gender and age, or other properties of a person that can be derived from visual appearance. Gathering anonymous knowledge about people from visual cues bears potential for many interesting applications, as for example in the area of human machine interfacing, targeted advertisement or video surveillance. Most related work investigates visual traits recognition from facial features of a person, with good recognition performance. Few systems have recently applied recognition on low resolution full-body images, which shows lower performance than the facial regions but already can deliver classification results even if no face is visible. Obviously full-body classification is more challenging, mainly due to large variations in body pose, clothing and occlusion. In our study we present an approach to human visual traits recognition, based on Histogram of oriented Gradients (HoG), colour features and Support Vector Machines (SVM). In this experimental study we focus on gender classification. Motivated by our application of real-time adaptive advertisement on public situated displays, and unlike previous works, we perform a thorough evaluation on much more comprehensive datasets that include hard cases like side- and back views. The extended annotations used in our evaluation will be published. We further show that a hierarchical classification scheme to disambiguate a person's directional orientation and additional colour features can increase recognition rates. Finally, we demonstrate that temporal integration of per-frame classification scores significantly improves the overall classification performance for tracked individuals and clearly outperforms current state-of-the-art accuracy for single images.
Geometry-aware Video Registration
(The Eurographics Association, 2010) Palma, Gianpalo; Callieri, Marco; Dellepiane, Matteo; Corsini, Massimiliano; Scopigno, Roberto; Reinhard Koch and Andreas Kolb and Christof Rezk-Salama
We present a new method for the accurate registration of video sequences of a real object over its dense triangular mesh. The goal is to obtain an accurate video-to-geometry registration to allow the bidirectional data transfer between the 3D model and the video using the perspective projection defined by the camera model. Our solution uses two different approaches: feature-based registration by KLT video tracking, and statistic-based registration by maximizing the Mutual Information (MI) between the gradient of the frame and the gradient of the rendering of the 3D model with some illumination related properties, such as surface normals and ambient occlusion. While the first approach allows a fast registration of short sequences with simple camera movements, the MI is used to correct the drift problem that KLT tracker produces over long sequences, due to the incremental tracking and the camera motion. We demonstrate, using synthetic sequences, that the alignment error obtained with our method is smaller than the one introduced by KLT, and we show the results of some interesting and challenging real sequences of objects of different sizes, acquired under different conditions.
Convex Relaxation for Grain Segmentation at Atomic Scale
(The Eurographics Association, 2010) Boerdgen, Markus; Berkels, Benjamin; Rumpf, Martin; Cremers, Daniel; Reinhard Koch and Andreas Kolb and Christof Rezk-Salama
Grains are material regions with different lattice orientation at atomic scale. They can be resolved on material surfaces with recent image acquisition technology. Simultaneously, new microscopic simulation tools allow to study mechanical models of grain structures. The robust and reliable identification and visualization of grain boundaries - in images both from simulation and from experiments - is of central importance in the field of material surface analysis. In this work, we compare a variety of variational approaches for grain boundary estimation from microscopy and simulation images. In particular, we show that grain boundary estimation can be solved by means of recently introduced convex relaxation techniques. These techniques allow to compute global solutions or solutions within a known bound of the optimum. Moreover, experimental results both on simulated and on transmission electron microscopy images confirm that the convex relaxation techniques provide significant improvements of the estimated grain boundaries over previously employed multiphase level set formulations.
Analysis of Structural Dependencies for the Automatic Visual Inspection of Wire Ropes
(The Eurographics Association, 2010) Haase, Daniel; Wacker, Esther-Sabrina; Schukat-Talamazzini, Ernst Günter; Denzler, Joachim; Reinhard Koch and Andreas Kolb and Christof Rezk-Salama
Automatic visual inspection is an arising field of research. Especially in security relevant applications, an automation of the inspection process would be a great benefit. For wire ropes, a first step is the acquisition of the curved surface with several cameras located all around the rope. Because most of the visible defects in such a rope are very inconspicuous, an automatic defect detection is a very challenging problem. As in general there is a lack of defective training data, most of the presented ideas for automatic rope inspection are embedded in a one-class classification framework. However, none of these methods makes use of the context information which results from the fact that all camera views image the same rope. In contrast to an individual analysis of each camera view, this work proposes the simultaneous analysis of all available camera views with the help of a vector autoregressive model. Moreover, various dependency analysis methods are used to give consideration to the regular rope structure and to deal with the high dimensionality of the problem. These dependencies are then used as constraints for the vector autoregressive model, which results in a sparse but powerful detection system. The proposed method is evaluated by using real wire rope data and the conducted experiments show that our approach clearly outperforms all previously presented methods.
Evaluation of Geometric Registration Methods for Using Spatial Augmented Reality in the Automotive Industry
(The Eurographics Association, 2010) Menk, Christoffer; Jundt, Eduard; Koch, Reinhard; Reinhard Koch and Andreas Kolb and Christof Rezk-Salama
Spatial augmented reality is especially interesting for the automotive industry, because in the production process of a car a lot of virtual content and also real objects are used. Therefore, the virtual content can be directly projected onto the real object to combine the advantages of the real and virtual world. One important issue for the usage of spatial augmented reality in automotive processes and applications is that the virtual content has to be projected with a very high accuracy onto the real object, because decisions are made on the basis of the projection. Therefore, we present in this article a new method for the evaluation of geometric registration techniques which align a projector to a real object. Additionally, we use this proposed method to evaluate existing geometric registration techniques. Furthermore, we present a new application where a projector is used to support the design process of a new car.
QEM-Filtering: A New Technique for Feature-Sensitive Terrain Mesh Simplification
(The Eurographics Association, 2010) Löffler, Falko; Schumann, Heidrun; Reinhard Koch and Andreas Kolb and Christof Rezk-Salama
Terrain simplification generates multi-resolution models, from which - traditionally - irregular or semi-regular triangulations are extracted to render a terrain at a suitable level of detail. Recent terrain simplification techniques, in contrast, rely on GPU-friendly regular grids and generate multiple resolutions by applying the filtering and sub-sampling paradigm. However, due to the smoothing and uniform sampling, these techniques sparsely approximate the terrain surface. Consequently, in order to guarantee a certain error threshold, considerably more triangles need to be rendered. In this paper, we present a novel feature-sensitive simplification technique. Our approach follows the aforementioned paradigm. The key idea is to maintain the regularity while recomputing the vertex positions by taking a specific error metric into account, namely the quadric error metric (QEM). Compared to previous approaches, we apply the paradigm to the grid of vertex-associated quadrics. From these we extract vertices of the new resolution by relying on quadric error minimization. We, thus, maintain the regular grid structure while preserving terrain features. Compared to methods, which are solely based on vertex-filtering and sub-sampling, our approach reduces the approximation error. As a consequence, we require fewer triangles, which improves the rendering performance.
Tuvok, an Architecture for Large Scale Volume Rendering
(The Eurographics Association, 2010) Fogal, Thomas; Krüger, Jens; Reinhard Koch and Andreas Kolb and Christof Rezk-Salama
In this paper we present the Tuvok architecture, a cross-platform open-source volume rendering system that delivers high quality, state of the art renderings at production level code quality. Due to its progressive rendering algorithm, Tuvok can interactively visualize arbitrarily large data sets even on low-end 32bit systems, though it can also take full advantage of high-end workstations with large amounts of memory and modern GPUs. To achieve this Tuvok uses an optimized out-of-core, bricked, level of detail data representation. From a software development perspective, Tuvok is composed of three independent components, a UI subsystem based on Qt, a rendering subsystem based on OpenGL and DirectX, and an IO subsystem. The IO subsystem not only handles the out-of-core data processing and paging but also includes support for many widely used file formats such as DICOM and ITK volumes. For rendering, Tuvok implements a wide variety of different rendering methods, ranging from 2D texture stack based approaches for low end hardware, to 3D slice based implementations and GPU based ray casters. All of these modes work with one- or multi-dimensional transfer functions, isosurface, and ClearView rendering modes. We also present ImageVis3D, a volume rendering application that uses the Tuvok subsystems. While these features may be found individually in other volume rendering packages, to our best knowledge this is the first open source system to deliver all of these capabilities at once.
Parallel View-Dependent Out-of-Core Progressive Meshes
(The Eurographics Association, 2010) Derzapf, Evgenij; Menzel, Nicolas; Guthe, Michael; Reinhard Koch and Andreas Kolb and Christof Rezk-Salama
The complexity of polygonal models is growing faster than the ability of graphics hardware to render them in real-time. If a scene contains many models and textures, it is often also not possible to store the entire geometry in the graphics memory. A common way to deal with such models is to use multiple levels of detail (LODs), which represent a model at different complexity levels. With view-dependent progressive meshes it is possible to render complex models in real time, but the whole progressive model must fit into graphics memory. To solve this problem out-of-core algorithms have to be used to load mesh data from external data devices. Hierarchical level of detail (HLOD) algorithms are a common solution for this problem, but they have numerous disadvantages. In this paper, we combine the advantages of view-dependent progressive meshes and HLODs by proposing a new algorithm for real-time view-dependent rendering of huge models. Using a spatial hierarchy we extend parallel view-dependent progressive meshes to support out-of-core rendering. In addition we present a compact data structure for progressive meshes, optimized for parallel GPU-processing and out-of-core memory management.
Multi-image Interpolation based on Graph-Cuts and Symmetric Optic Flow
(The Eurographics Association, 2010) Linz, Christian; Lipski, Christian; Magnor, Marcus A.; Reinhard Koch and Andreas Kolb and Christof Rezk-Salama
Multi-image interpolation in space and time has recently received considerable attention. Typically, the interpolated image is synthesized by adaptively blending several forward-warped images. Blending itself is a low-pass filtering operation: the interpolated images are prone to blurring, even if correspondences are perfect. Furthermore, ghosting artifacts appear as soon as the underlying correspondence fields are imperfect. We address both issues and propose a multi-image interpolation algorithm that avoids blending. Instead, we cast multi-image interpolation as a labeling problem and decide for each pixel in the synthesized view from which input image to sample. Combined with a symmetrical long-range optical flow formulation for correspondence field estimation, our approach yields crisp interpolated images without ghosting artifacts.
Massively Parallel Multiclass Object Recognition
(The Eurographics Association, 2010) Sedding, Helmut; Deger, Ferdinand; Dammertz, Holger; Bouecke, Jan; Lensch, Hendrik P. A.; Reinhard Koch and Andreas Kolb and Christof Rezk-Salama
We present a massively parallel object recognition system based on a cortex-like structure. Due to its nature, this general, biologically motivated system can be parallelized efficiently on recent many-core graphics processing units (GPU). By implementing the entire pipeline on the GPU, by rigorously optimizing memory bandwidth and by minimizing branch divergence, we achieve significant speedup compared to both recent CPU as well as GPU implementations for reasonably sized feature dictionaries. We demonstrate an interactive application even on a less powerful laptop which is able to classify webcam images and to learn novel categories in real time.
Hardware Accelerated 3D Mesh Painting
(The Eurographics Association, 2010) Schärfig, Randolf; Hormann, Kai; Reinhard Koch and Andreas Kolb and Christof Rezk-Salama
In this paper we present a new algorithm for interactively painting onto 3D meshes that exploits recent advances of GPU technology. As the user moves a brush over the 3D mesh, its paint pattern is projected onto the 3D geometry at the current viewing angle and copied to the corresponding region in the object's texture atlas. Both operations are realized on the GPU, with the advantage that all data resides in the fast GPU memory, which in turn leads to high frame rates. A main feature of our approach is the handling of seams. Whenever the brush overlaps two or more patches, this situation is detected and the paint pattern is copied correctly to the corresponding texture charts. In this way the operation of the projection into the texture atlas is completely reduced to a single texture lookup. The performance is independent of the resolution of both the brush and the texture atlas as well as the number of mesh triangles.
Topology-Caching for Dynamic Particle Volume Raycasting
(The Eurographics Association, 2010) Orthmann, Jens; Keller, Maik; Kolb, Andreas; Reinhard Koch and Andreas Kolb and Christof Rezk-Salama
In this paper we present a volume rendering technique for the ad-hoc visualization of interactive particle systems. We focus on methods for an efficient spatial caching (topology caching) of particles when applying a raycasting approach. Thus, we get a fast reconstruction of the scalar field which is defined by the particles' entities. The node-cache allows for efficient caching and pre-fetching of a subset of the octree nodes. The influence-cache provides fast access to all particles which contribute to a specific node including level-of-detail particles. Finally, the introduced slab-cache allows for efficient volume rendering and gradient computation. Our algorithms are completely built and managed on the GPU and interactive frame rates for up to several 105 particles are achieved.
The Minimal Bounding Volume Hierarchy
(The Eurographics Association, 2010) Bauszat, Pablo; Eisemann, Martin; Magnor, Marcus; Reinhard Koch and Andreas Kolb and Christof Rezk-Salama
Bounding volume hierarchies (BVH) are a commonly used method for speeding up ray tracing. Even though the memory footprint of a BVH is relatively low compared to other acceleration data structures, they still can consume a large amount of memory for complex scenes and exceed the memory bounds of the host system. This can lead to a tremendous performance decrease on the order of several magnitudes. In this paper we present a novel scheme for construction and storage of BVHs that can reduce the memory consumption to less than 1% of a standard BVH. We show that our representation, which uses only 2 bits per node, is the smallest possible representation on a per node basis that does not produce empty space deadlocks. Our data structure, called the Minimal Bounding Volume Hierarchy (MVH) reduces the memory requirements in two important ways: using implicit indexing and preset surface reduction factors. Obviously, this scheme has a non-negligible computational overhead, but this overhead can be compensated to a large degree by shooting larger ray bundles instead of single rays, using a simpler intersection scheme and a two-level representation of the hierarchy. These measure enable interactive ray tracing performance without the necessity to rely on out-of-core techniques that would be inevitable for a standard BVH.
Interactive Multi-View Facade Image Editing
(The Eurographics Association, 2010) Musialski, Przemyslaw; Luksch, Christian; Schwärzler, Michael; Buchetics, Matthias; Maierhofer, Stefan; Purgathofer, Werner; Reinhard Koch and Andreas Kolb and Christof Rezk-Salama
We propose a system for generating high-quality approximated façade ortho-textures based on a set of perspective source photographs taken by a consumer hand-held camera. Our approach is to sample a combined orthographic approximation over the facade-plane from the input photos. In order to avoid kinks and seams which may occur on transitions between different source images, we introduce color adjustment and gradient domain stitching by solving a Poisson equation in real-time. In order to add maximum control on the one hand and easy interaction on the other, we provide several editing interactions allowing for user-guided post-processing.
Geometric Reconstruction of the Ostium of Cerebral Aneurysms
(The Eurographics Association, 2010) Neugebauer, Mathias; Diehl, Volker; Skalej, Martin; Preim, Bernhard; Reinhard Koch and Andreas Kolb and Christof Rezk-Salama
Polygonal 3D-reconstructions of cerebral aneurysms, combined with simulated or measured flow data provide important information for medical research, risk assessment and therapy planning. Landmarks, orientation axis, and a subdivision into functional unities, support the purposeful exploration of this complex data. The ostium, the area of inflow into the aneurysm, is the reference structure for various landmarks, axis and the initial subdivision into aneurysm's body and parent vessel.We present an approach to automatically extract important landmarks and geometrically reconstruct the ostium. Our method was successfully applied to various types of saccular aneurysms. These results were discussed with radiology experts. Our approach was considered as useful to reduce interpersonal variance in the ostium determination and forms a basis for subsequent quantification and exploration.
Proxy-Guided Texture Synthesis for Rendering Natural Scenes
(The Eurographics Association, 2010) Bonneel, Nicolas; Panne, Michiel van de; Lefebvre, Sylvain; Drettakis, George; Reinhard Koch and Andreas Kolb and Christof Rezk-Salama
Landscapes and other natural scenes are easy to photograph but difficult to model and render. We present a proxy-guided pipeline which allows for simple 3D proxy geometry to be rendered with the rich visual detail found in a suitably pre-annotated example image. This greatly simplifies the geometric modeling and texture mapping of such scenes. Our method renders at near-interactive rates and is designed by carefully adapting guidancebased texture synthesis to our goals. A guidance-map synthesis step is used to obtain silhouettes and borders that have the same rich detail as the source photo, using a Chamfer distance metric as a principled way of dealing with discrete texture labels. We adapt an efficient parallel approach to the challenging guided synthesis step we require, providing a fast and scalable solution. We provide a solution for local temporal coherence, by introducing a reprojection algorithm, which reuses earlier synthesis results when feasible, as measured by a distortion metric. Our method allows for the consistent integration of standard CG elements with the texture-synthesized elements. We demonstrate near-interactive camera motion and landscape editing on a number of examples.
Ultrasound Painting of Liver Vascular Tree
(The Eurographics Association, 2010) Birkeland, Asmund; Viola, Ivan; Reinhard Koch and Andreas Kolb and Christof Rezk-Salama
In treatment planning and surgical interventions, physicians and surgeons need information about the spatial extent of specific features and the surrounding structures. Previous techniques for extracting features, based on magnetic resonance imaging and computed tomography scans, can be slow and cumbersome and are rarely used by doctors. In this paper we will present a novel approach to extract features from tracked 2D ultrasound, in particular hypo-echoic regions such as blood vessels. Features are extracted during live examinations, removing the need for slow and cumbersome post-scan processes and interaction is based on the natural interaction techniques used by doctors during the examination. The ultrasound probe is utilized as a 3D brush, painting features in a 3D environment. The painting occurs during a regular examination, producing little extra interaction from the doctor. We will introduce a novel approach to extract hypo-echoic regions from an ultrasound image and track the regions from frame to frame. 3D models are then generated by storing the outline of the region as a 3D point cloud. Automatically detecting branching, this technique can handle complex structures, such as liver vessel trees, and track multiple regions simultaneously. During the examination, the point cloud is triangulated in real-time, enabling the doctor to examine the results live and discard areas which are unsatisfactory. To enable modifications of the extracted 3D models, we present how the ultrasound probe can be used as a interaction tool for fast point cloud editing.

Browse

Browsing VMV: Vision, Modeling, and Visualization by Issue Date

Results Per Page

Sort Options