DC-Seg: Hierarchical Image Segmentation    
"Discrete-Continuous Gradient Orientation Estimation for Faster Image Segmentation", Michael Donoser and Dieter Schmalstieg, Proceedings of Conference on Computer Vision and Pattern Recognition (CVPR), 2014
 

In this work, we propose an image segmentation method, that returns a hierarchy of segments with differing granularity. We propose an efficient discrete-continuous optimization of oriented gradient signals, that are passed to an oriented watershed transformation to build a hierarchical segment tree as output. We are able to provide segmentation performance competitive to state-of-the-art (even without any additional spectral analysis) while reducing computation time by a factor of 40. Additionally, since we do not have to apply a spectral analysis, we are able to reduce the memory profile by a factor of 10. In such a way, our segmentation method allows to provide useful input to various computer vision applications in short computation time.
Code and more details can be found on the DC-Seg: Image Segmentation page.

Contact: Michael Donoser (michael.donoser(at)tugraz.at)


Embedded Ferns: Discriminative Feature-to-Point Matching   
"Discriminative Feature-to-Point Matching in Image-Based Localization", Michael Donoser and Dieter Schmalstieg, Proceedings of Conference on Computer Vision and Pattern Recognition (CVPR), 2014
 

In this work we approach the problem of image-based localization, i.\,e.~how to infer an accurate camera pose from a given image within a known 3D world. While the prevalent approach to image-based localization is to match interest points detected in the query image to a sparse 3D point cloud representing the world using using nearest neighbour analysis, we define this correspondence finding problem as a classification task. We propose an extension of the random fern principle, denoted as the embedded random fern, by projecting features to fern-specific embedding spaces, which yields improved matching rates in short runtime.
Code and more details can be found on the Discriminative Feature-to-Point Matching page.

Contact: Michael Donoser (michael.donoser(at)tugraz.at)


Diffusion Processes for Retrieval   
"Diffusion Processes for Retrieval", Michael Donoser and Horst Bischof, Proceedings of Conference on Computer Vision and Pattern Recognition (CVPR), 2013
 

In this work we revisit diffusion processes on affinity graphs for capturing the intrinsic manifold structure defined by pairwise affinity matrices. Based on our observations on related work, we are able to derive a generic framework for diffusion processes in the scope of retrieval applications, where the related work represents specific instances of our generic formulation. Experiments demonstrate applicability of our diffusion framework for retrieval, e.g. achieving a 100% bullseye score on the popular MPEG-7 shape retrieval data set.
Code and more details can be found on the Diffusion Processes page.

Contact: Michael Donoser (michael.donoser(at)tugraz.at)


Replicator Graph Clustering   
"Replicator Graph Clustering", Michael Donoser, Proceedings of British Conference on Computer Vision (BMVC), 2013
 

In this work we introduce an efficient, effective and scalable clustering method denoted as Replicator Graph Clustering. Our method takes measures of similarity between pairs of data points (i.e. an affinity matrix) as input and identifies a set of clusters and unique cluster assignments in a fully unsupervised manner, where the cluster granularity is adaptable by a single parameter. We provide results in three subsequent steps: (a) propagating affinities by finding personalized evolutionary stable strategies of noncooperative games (b) building a mutual k-nearest neighbor graph representing the underlying manifold and (c) applying a graph based clustering strategy which identifies the final clusters. Individual steps have low computational complexity which leads to a highly efficient clustering method, scaling well with an increasing number of data points. Experimental evaluation demonstrates competitive performance to state-of-the-art in several application fields.
Code and more details can be found on the Replicator Graph Clustering page.

Contact: Michael Donoser (michael.donoser(at)tugraz.at)


Optimizing 1-Nearest Prototype Classifiers   
"Optimizing 1-Nearest Prototype Classifiers", Paul Wohlhart, Martin Kostinger, Michael Donoser, Peter M. Roth, Horst Bischof, Proceedings of Conference on Computer Vision and Pattern Recognition (CVPR), 2013
1-NN Prototypes
 

In this work we advocate simple nearest neighbor classifiers, which have several beneficial properties, such as low complexity and inherent multi-class handling. However, they have a runtime linear in the size of the database. Recent related work represents data samples by assigning them to a set of prototypes that partition the input feature space and afterwards applies linear classifiers on top of this representation to approximate decision boundaries locally linear. In this paper, we go a step beyond these approaches and purely focus on 1-nearest prototype classification, where we propose a novel algorithm for deriving optimal prototypes in a discriminative manner from the training samples. Our method is implicitly multi-class capable, parameter free, avoids noise overfitting and, since during testing only comparisons to the derived prototypes are required, highly efficient. Experiments demonstrate that we are able to outperform related locally linear methods, while even getting close to the results of more complex classifiers.

Contact: Paul Wohlhart (wohlhart(at)icg.tugraz.at) Michael Donoser (michael.donoser(at)tugraz.at)


Hough Regions
"Hough Regions for Joining Instance Localization and Segmentation", Hayko Riemenschneider, Sabine Sternig, Michael Donoser, Peter M. Roth and Horst Bischof, Proceedings of European Conference on Computer Vision (ECCV), 2012
Hough Regions
 

In this work we propose a framework which jointly optimizes for object detection and accurate segmentations per instance. Our novel approach is attachable to any of the available generalized Hough voting methods. We introduce Hough Regions by formulating the problem of Hough space analysis as Bayesian labeling of a random field. In such a way we bypass the parameter sensitive non-maximum suppression that is required in related methods. Experimental evaluation demonstrates that our method is inherently able to handle overlapping instances and an increased range of articulations, aspect ratios and scales.

Contact: Hayko Riemenschneider (hayko(at)vision.ee.ethz.ch), Michael Donoser (michael.donoser(at)tugraz.at)


Kernel Functions for Graph Matching
"Learning Edge-Specific Kernel Functions For Pairwise Graph Matching", Michael Donoser, Martin Urschler and Horst Bischof, Proceedings of British Conference on Computer Vision (BMVC), 2012
Kernels for Graph Matching
 

In this work we consider the pairwise graph matching problem of finding correspondences between two point sets using unary and pairwise potentials. Our novel approach learns edge-specific kernels for pairs of nodes from training data. Assuming that the setting of graph matching is a priori known, the learned kernel functions allow to significantly improve results in comparison to general graph matching. Experiments demonstrate the possible improvements of our proposed method.

Contact: Michael Donoser (michael.donoser(at)tugraz.at)


Robust Planar Target Tracking
"Robust Planar Target Tracking and Pose Estimation from a Single Concavity", Michael Donoser, Peter Kontschieder and Horst Bischof, Proceedings of International Symposium of Mixed and Augmented Reality (ISMAR), 2011
 

In this work we introduce a novel real-time method to track weakly textured planar objects and to simultaneously estimate their 3D pose. The basic idea is to adapt the classic tracking-by-detection approach, which seeks for the object to be tracked independently in each frame, for tracking non-textured objects. Experiments show that we obtain robust tracking results with accurate poses on various challenging image sequences.
More details and videos can be found on the Robust Planar Tracking page.

Contact: Michael Donoser (michael.donoser(at)tugraz.at)


Efficient Shape Based Object Category Localization
"Using Partial Edge Contour Matches for Efficient Object Category Localization", Hayko Riemenschneider, Michael Donoser and Horst Bischof, Proceedings of European Conference on Computer Vision (ECCV), 2010
 

In this work we propose a method for object category localization by partially matching edge contours to a single shape prototype of the category. Previous work in this area either relies on piecewise contour approximations, requires meaningful supervised decompositions, or matches coarse shape-based descriptions at local interest points. Our method avoids error-prone pre-processing steps by using all obtained edges in a partial contour matching setting. The matched fragments are efficiently summarized and aggregated to form location hypotheses. The effciency and accuracy of our edge fragment based voting step yields high quality hypotheses in low computation time. The experimental evaluation achieves excellent state-of-the-art performance in the hypotheses voting stage and yields competitive results on challenging datasets like ETHZ and INRIA horses.
More details and results can be found on the Efficient Object Category Localization page.

Contact: Hayko Riemenschneider (hayko(at)icg.tugraz.at) or Michael Donoser (michael.donoser(at)tugraz.at)


Edge Detection for Object Localization
"Linked Edges as Stable Region Boundaries", Michael Donoser, Hayko Riemenschneider and Horst Bischof, Proceedings of Conference on Computer Vision and Pattern Recognition (CVPR), 2010
 

In this work we introduce a method to detect edges as the most stable region boundaries designed for the task of object localization. In contrast to common edge detection algorithms as Canny, which only analyze local discontinuities in image brightness, our method integrates mid-level information by analyzing regions that support the local gradient magnitudes. We use a component tree where every node contains a single connected region obtained from thresholding the gradient magnitude image. Edges in the tree are defined by an inclusion relationship between nested regions in different levels of the tree. Region boundaries which are similar in shape (i.e. have a low chamfer distance) across several levels of the tree are included in the final result. Since the component tree can be calculated in quasi-linear time and chamfer matching between nodes in the component tree is reduced to analysis of the distance transformation, results are obtained in an efficient manner. The proposed detection algorithm labels all identified edges during calculation, thus avoiding the cumbersome post-processing of connecting and labeling edge responses. We evaluate our method on two reference data sets and demonstrate improved performance for shape prototype based localization of objects in images.
Code, detection results and more details can be found on the Edge Detection page.

Contact: Michael Donoser (michael.donoser(at)tugraz.at) or Hayko Riemenschneider (hayko(at)icg.tugraz.at)


Saliency Driven Unsupervised Segmentation
"Saliency Driven Total Variation Segmentation", Michael Donoser, Martin Urschler, Martin Hirzer and Horst Bischof, Proceedings of International Conference on Computer Vision (ICCV), 2009
 

In this work we consider the problem of unsupervised segmentation. We poposed a general concept, which is based on the underlying idea of segmenting the input image several times, each time focussing on a different salient part of the image and to subsequently merge all obtained results into one composite segmentation. Any saliency detector or efficient color space clustering apporahc can be used to provide the salient regions. Each salient region then serves as an independent initialization for a figure/ground segmentation. Segmentation is done by minimizing a convex energy functional based on weighted total variation leading to a global optimal solution. Each salient region provides an accurate figure/ground segmentation highlighting different parts of the image. These highly redundant results are combined into one composite segmentation by analyzing local segmentation certainty. We demonstrate the high quality of our method on the well-known Berkeley segmentation database.
Code, examplary results and more details can be found on the Unsupervised Segmentation page.

Contact: Michael Donoser (michael.donoser(at)tugraz.at) or Martin Urschler (urschler(at)icg.tugraz.at)


Tracking by Structure Tensor Analysis
"Object Tracking by Structure Tensor Analysis", Michael Donoser, Stefan Kluckner and Horst Bischof, In In Proceedings International Conference on Pattern Recognition (ICPR), 2010
 

In this work we present an object tracking approach that is based on the well-known structure tensor. We show that the generalized structure tensor is a powerful descriptor which can be calculated in constant time by integral images. We furthermore describe an approximation scheme which allows comparison of multiple structure tensors in an efficient manner in Euclidean space in contrast to more computationally demanding Riemannian Manifold distances. Experimental evaluation proves the applicability for the task of object tracking demonstrating improved performance. As a general descriptor strutcure tensors are directly applicable in any e.g. particle filter based tracking framework.
Code, examplary results and more details can be found on the Tracking by Structure Tensors page.

Contact: Michael Donoser (michael.donoser(at)tugraz.at) or Stefan Kluckner (kluckner(at)icg.tugraz.at)


Beyond Pairwise Shape Similarity Analysis
"Beyond Pairwise Shape Similarity Analysis", Peter Kontschieder, Michael Donoser and Horst Bischof, Proceedings of Asian Conference on Computer Vision (ACCV), 2009
 

In this work we consider two major applications of shape matching algorithms: (a) query-by-example, i.e. retrieving the most similar shapes from a database and (b) finding clusters of shapes, each represented by a single prototype. Our approach goes beyond pairwise shape similarity analysis by considering the underlying structure of the shape manifold, which is estimated from the shape similarity scores between all the shapes within a database. We propose a modified mutual kNN graph as the underlying representation and demonstrate its performance for the task of shape retrieval. We further describe an efficient, unsupervised clustering method which uses the modified mutual kNN graph for initialization. Experimental evaluation proves the applicability of our method, e.g. by achieving a retrieval score of 93.40% on the well known MPEG-7 database.
Code and more details can be found on the Shape Retrieval page.

Contact: Peter Kontschieder (kontschieder(at)tugraz.at) or Michael Donoser (michael.donoser(at)tugraz.at)


Maximally Stable Extremal Region (MSER) Tracking
"Efficient Maximally Stable Extremal Region (MSER) Tracking", Michael Donoser and Horst Bischof Proceedings of Conference on Computer Vision and Pattern Recognition (CVPR), 2006
 

In this work we consider the problem of tracking interest points in a highly repetetive way through image sequences. Maximally Stable Extremal Regions (MSERs) have shown to be one of the best interest detectors in computer vision, and we focus on tracking these regions through sequences to improve detection repeatability. We use the component tree as an efficient data structure, which allows the calculation of MSERs in quasi-linear time. It is demonstrated that the tree is able to manage the required data for tracking. We show that by means of MSER tracking the computational time for the detection of single MSERs can be improved by a factor of 4 to 10. Using a weighted feature vector for data association improves the tracking stability. Furthermore, the component tree enables backward tracking which further improves the robustness.
More details can be found on the MSER Tracking page.

Contact: Michael Donoser (michael.donoser(at)tugraz.at)


Semi-Supervised Stability Guided 2D and 3D Segmentation
"ROI-SEG: Unsupervised Color Segmentation by Combining Differently Focused Sub Results", Michael Donoser and Horst Bischof Proceedings of Conference on Computer Vision and Pattern Recognition (CVPR), 2007
 

In this work we consider the problem of semi-supervised segmentation, i.e. given a manually initialization of the area of interest we provide accurate figure/ground segmentations. The core idea is to exploit the properties of the Maximally Stable Extremal Region (MSER) detector (frequently used for interest region detection in computer vision) for the purpose of segmentation. Modeling the color/texture distribution of the initialization provided by the user returns local figure/ground probabilities, which are passed to the MSER detector. Both modeling and MSER detection is done in an efficient manner yielding results in low computation time. We show single image and video sequence segmentation results which prove the applicability of the concept. Additionally, the possible extension to 3D segmentation is shown by an application that analyzes the 3D microstructure of a sheet of paper.
Code and more details can be found on the Semi-Automatic Segmentation page.

Contact: Michael Donoser (michael.donoser(at)tugraz.at)


Semantic Classification in Large-Scale Aerial Imagery
Stefan Kluckner and Horst Bischof
 

This work introduces an efficient classification pipeline, which provides an accurate semantic interpretation of urban environments by using redundant scene observations. The image-based method integrates both appearance and height data to classify single aerial images. Given the initial classification of highly overlapping images, a projection to a common orthographic 3D world coordinate system provides redundant observations from multiple viewpoints and enables a semantic interpretation of large-scale urban environments. In the experimental evaluation we investigate how the use of redundancy influences the accuracy in terms of correctly classified pixels for object classes like building, tree, grass, street and water areas. Moreover, we exploit an efficient yet continuous formulation of the Potts model to obtain a consistent labeling of the pixels in the orthographic view.
Code, classification results and more details can be found on the Semantic Classification in Large-Scale Aerial Imagery page.

Contact: Stefan Kluckner(kluckner(at)icg.tugraz.at)


Facade Segmentation and Separation
"Unsupervised Facade Segmentation using Repetitive Patterns", Andreas Wendel, Michael Donoser, and Horst Bischof, Proceedings of the DAGM Conference, Darmstadt (Germany), 2010
Unsupervised Facade Segmentation using Repetitive Patterns
 

We introduce a novel approach for separating and segmenting individual facades from streetside images. Our algorithm incorporates prior knowledge about arbitrarily shaped repetitive regions which are detected using intensity profi le descriptors and a voting-based matcher. In the experiments we compare our approach to extended state-of-the-art matching approaches using more than 600 challenging streetside images, including different building styles and various occlusions. Our algorithm outperforms these approaches and allows to correctly separate 94% of the facades. Pixel-wise comparison to our ground-truth yields a segmentation accuracy of 85%. According to these results our work is an important contribution to fully automatic building reconstruction.
Please also watch our video which demonstrates the algorithm and shows some results.

Contact: Andreas Wendel (wendel(at)icg.tugraz.at) or Michael Donoser (michael.donoser(at)tugraz.at)


Efficient Partial Shape Matching
"Efficient Partial Shape Matching of Outer Contours", Michael Donoser, Hayko Riemenschneider and Horst Bischof, Proceedings of Asian Conference on Computer Vision (ACCV), 2009
 

In this work we consider the problem of efficiently matching binary silhouettes. We use sampled points from the silhouette as a shape representation. The sampled points can be ordered which in turn allows to formulate the matching step as an order-preserving assignment problem. We propose an angle descriptor between shape chords combining the advantages of global and local shape description. An efficient integral image based implementation of the matching step is introduced which allows detecting partial matches an order of magnitude faster than comparable methods. We further show how the proposed algorithm is used to calculate a global optimal Pareto frontier to de ne a partial similarity measure between shapes. We demonstrate the applicability of the approach for shape retrieval on standard shape databases like MPEG-7 and for tracking purposes.
More details can be found on the Shape Matching page.

Contact: Michael Donoser (michael.donoser(at)tugraz.at) or Hayko Riemenschneider (hayko(at)icg.tugraz.at)


Tracking by Repeated Figure/Ground Segmentation
"Fast Non-Rigid Object Boundary Tracking", Michael Donoser and Horst Bischof, Proceedings of British Machine Vision Conference (BMVC), 2009
Tracking as Repeated Figure/Ground Segmentation
 

In this work we consider the problem of tracking an object in a robust way through an image sequence and at the same tome providing accurate object outlines in each frame. We describe a tracking by repeated figure/ground segmentation method, which has to be initialized by drawing a bounding box in the first frame. We first calculate edge responses on efficiently calculated color probability maps in an object-specific Fisher color space. The estimated edge maps are used in a probabilistic particle filtering framework hypothesizing rigid transformations for initializing an active contour model. Our approach provides accurate object segmentations in every frame of the sequence. We mainly apply the tracker for tracking faces and hands, e.g. to steer the mouse cursor by video analysis.

Contact: Michael Donoser (michael.donoser(at)tugraz.at)


Using Web Search Engines to Improve Text Recognition
"Using Web Search Engines to Improve Text Recognition", Michael Donoser, Silke Wagner and Horst Bischof, Proceedings of International Conference on Pattern Recognition (ICPR), 2008
Using Web Serach Engines
 

In this work we describe a method for post-processing of text recognition results. Most of the approaches in this field compare provided hypotheses to a dictionary or exploit learned statitistics e.g. about syllables-cooccurences. The main idea in our work is to use web search engine results to verify the hypotheses and to exploit contextual information, which is present in search results. Verification is done on two levels of detail (word and sentence level) which both improve the overall text recognition performance. Experimental evaluations demonstrate that even based on a low-quality single character recognition method the proposed web search engine extension enables reasonable text recognition results.

Contact: Michael Donoser (michael.donoser(at)tugraz.at) or Silke Wagner (wagnersi(at)edu.uni-graz.at)


Automated 3D Structure Image Acquisition
"Automated serial sectioning applied to 3D paper structure analysis", Mario Wiltsche, Michael Donoser, Johannes Kritzinger, Wolfgang Bauer
 

In this work we introduce a novel 3D image acquisition system which enables 3D digitization of embedded materials with high resolution, sufficient sample size and moderate operating time and costs. A prototype which combines serial sectioning and light microscopy was designed and built. The design of the developed prototype is shown below. The approach allows 3D digitization in a fully automated way without any need for user interaction. The applicability of the prototype was mainly shown for digitization of paper samples, but in general the methods can be used for digitizing various kinds of materials, thus applications in the field of medical histology, biology or polymer industry seem to be very promising.
More details can be found on the 3D Image Acquisition page.

Contact: Michael Donoser (michael.donoser(at)tugraz.at)