List of Publications
Copyright notice
The list of my publication as PDF. Please use this bibliography file [bibtex] when you reference my work.
 [75]
 Walter G. Kropatsch, Nicole M. Artner, Yll Haxhimusa, and Xiaoyi Jiang (eds). Special issue on graphbased representations in pattern recognition. Pattern Recognition, Special Issue of GbRPR 2013. 2014.
 [74]

Marco Augustin, Yll Haxhimusa, Wolfgang Busch, and Walter G. Kropatsch.
Imagebased phenotyping of the mature arabidopsis shoot system.
In Computer Vision Problems in Plant Phenotyping, In conjuction
with European Computer Vision Conference, 2014.
The imagebased phenotyping of mature plants faces several challenges from the image acquisition to the determination of quanti tative characteristics describing their appearance. In this work a frame work to extract geometrical and topological traits of 2D images of mature Arabidopsis thaliana is proposed. The phenotyping pipeline recovers the realistic branching architecture of dried and attened plants in two steps. In the first step, a tracing approach is used for the extraction of centerline segments of the plant. In the second step, a hierarchical reconstruction is done to group the segments according to continuity principles. This paper covers an overview of the relevant processing steps along the pro posed pipeline and provides an insight into the image acquisition as well as into the most relevant results from the evaluation process.
 [73]
 Walter G. Kropatsch, Nicole M. Artner, Yll Haxhimusa, and Xiaoyi Jiang (eds). GraphBased Representations in Pattern Recognition  9th IAPRTC15 International Workshop, GbRPR 2013, Vienna, Austria, May 1517, 2013. Proceedings, volume 7877 of Lecture Notes in Computer Science. Springer, 2013.
 [72]

Samuel de Sousa, Yll Haxhimusa, and Walter G. Kropatsch.
Estimation of distribution algorithm for the maxcut problem.
In Walter G. Kropatsch, Nicole M. Artner, Yll Haxhimusa, and Xiaoyi
Jiang, editors, GraphBased Representations in Pattern Recognition  9th
IAPRTC15 International Workshop, GbRPR 2013, Vienna, Austria, May 1517,
2013. Proceedings, volume 7877 of Lecture Notes in Computer Science,
pages 244253. Springer, 2013.
In this paper, we investigate the MaxCut problem and propose a probabilistic heuristic to address its weighted version. Our approach is based on the Estimation of Distribution Algorithm (EDA) that creates a population of individuals and at each generation it evolves aiming at the global maximum. We have applied the MaxCut problem on the image segmentation task and defined the edges weights as the L2 norm between the RGB values of the nodes. The main goal of this paper is to introduce a heuristic for MaxCut and to investigate how it can be applied in the segmentation context.
 [71]
 Esther Antunez, Yll Haxhimusa, Rebeca Marfil amd Walter G. Kropatsch, and Antonio Bandera. Artificial visual attention using combinatorial pyramids. In Jose García Rodriguez and Miguel Cazorla (eds), editors, Robotics and Vision: Technologies for Machine Learning and Vision Applications, pages 439457. IGI Global, 2012.
 [70]
 Yll Haxhimusa. Interactive labeling of image segmentation hierarchies. Statistiche Woche, pages 7172, Vienna, Austria, 2012.
 [69]

Philip Limbeck, Walter G. Kropatsch, and Yll Haxhimusa.
Semiautomatic Tracking of Markers in Facial Palsy.
In Yoshimitsu Aoki Hideo Saito (eds), editor, ICPR 2012, In the
Proceedings of the 21st Internatinal Conference in Pattern Recognition ICPR
2012, Tsukuba, Japan, pages 6972. IEEE Comp. Soc., 2012.
We introduce a semiautomatic tracking method that can be utilized for the analysis of facial markers in the medical condition of facial palsy. Tracking of markers will help medical physicians in evaluating this medical condition quantitatively. We use particle filtering to track markers towards measuring distances needed to evaluate the degree of facial palsy. We show that by employing tracking methods, the analysis time is reduced without losing the high accuracy of the results.
 [68]

Gayane Shalunts, Yll Haxhimusa, and Robert Sablatnig.
Segmentation of building facade domes.
In Luis Álvarez, Marta Mejail, Luis Gómez, and Julio C.
Jacobo, editors, Progress in Pattern Recognition, Image Analysis,
Computer Vision, and Applications  17th Iberoamerican Congress, CIARP 2012,
Buenos Aires, Argentina, September 36, 2012. Proceedings, volume 7441 of
Lecture Notes in Computer Science, pages 324331. Springer, 2012.
[ http ]
Domes are architectural structural elements typical for ecclesiastical and secular grand buildings, like churches, mosques, palaces, capitols and city halls. The current paper targets the problem of segmentation of domes within the framework of architectural style classification of building facades. We perform segmentation of building facade domes by combining bilateral symmetry detection, graphbased segmentation approaches and image analysis and processing technics into a single method. Our algorithm achieves good segmentation results on buildings belonging to variety of architectural styles, such as Renaissanse, NeoRenaissance, Baroque, NeoBaroque, Neoclassical and Islamic.
 [67]

Georg Zankl, Yll Haxhimusa, and Adrian Ion.
Interactive labeling of image segmentation hierarchies.
In Axel Pinz, Thomas Pock, Horst Bischof, and Franz Leberl, editors,
Pattern Recognition  Joint 34th DAGM and 36th OAGM Symposium, Graz,
Austria, August 2831, 2012. Proceedings, volume 7476 of Lecture Notes
in Computer Science, pages 1120. Springer, 2012.
[ http ]
We study the task of interactive semantic labeling of a segmentation hierarchy. To this end we propose a framework interleaving two components: an automatic labeling step, based on a Conditional Random Field whose dependencies are defined by the inclusion tree of the segmentation hierarchy, and an interaction step that integrates incremental input from a human user. Evaluated on two distinct datasets, the proposed interactive approach efficiently integrates human interventions and illustrates the advantages of structured prediction in an interactive framework.
 [66]

Gayane Shalunts, Yll Haxhimusa, and Robert Sablatnig.
Architectural style classification of domes.
In George Bebis, Richard Boyle, Bahram Parvin, Darko Koracin,
Charless Fowlkes, Sen Wang, MinHyung Choi, Stephan Mantler, Jürgen P.
Schulze, Daniel Acevedo, Klaus Mueller, and Michael E. Papka, editors,
Advances in Visual Computing  8th International Symposium, ISVC 2012,
Rethymnon, Crete, Greece, July 1618, 2012, Revised Selected Papers, Part
II, volume 7432 of Lecture Notes in Computer Science, pages 420429.
Springer, 2012.
[ http ]
Domes are architectural structural elements characteristic for ecclesiastical and secular monumental buildings, like churches, basilicas, mosques, capitols and city halls. In the scope of building facade architectural style classification the current paper addresses the problem of architectural style classification of facade domes. Building facade classification by architectural styles is achieved by classification and voting of separate architectural elements, like domes, windows, towers, etc. Typical forms of the structural elements bear the signature of each architectural style. Our approach classifies domes of three architectural styles  Renaissance, Russian and Islamic. We present a threestep approach, which in the first step analyzes the height and width of the dome for the identification of Islamic saucer domes, in the second step detects golden color in YCbCr color space to determine Russian golden onion domes and in the third step performs classification based on dome shapes, using clustering and learning of local features. Thus we combine three features  the relation of dome width and height, color and shape, in a single methodology to achieve high classification rate.
 [65]
 Iris van Rooij, Yll Haxhimusa, Zygmunt Pizlo, and Georg Gottlob. Computer Science & Problem Solving: New Foundations (Dagstuhl Seminar 11351). Dagstuhl Reports, 1(8):96124, 2011. [ DOI  http ]
 [64]

Adrian Ion, Yll Haxhimusa, and Walter G. Kropatsch.
Skewed coordinate system for dense point correspondences inside
articulated shapes.
Technical Report PRIPTR126, Pattern Recognition and Image
Processing Lab, Institute of Computer Aided Automation, Vienna University of
Technology, 2011.
[ .pdf 
.pdf ]
This paper considers using a nonrigid coordinate system to find corresponding points in different poses of the same articulated 2D shape. The shapecentered coordinate system is mapped on top of the eccentricity transform of the shape, which uses maximal geodesic distances and is bounded under articulation. The isolines of the eccentricity transform are used as one of the coordinates, the radiallike, and the other one, the angularlike, is stretched to compensate for changes in the widths of parts. The polarlike coordinate system is first computed on interpixel isolines and then mapped to the pixels. The angularlike coordinates are aligned using the 1D signals of the eccentricity values along the boundaries of the two shapes. Correspondences between points are established by minimizing the difference of their coordinates. Detecting failed correspondences is done using an adaptive threshold which adjusts to the changing local variation of the coordinates. Experimental results are shown on a set of hand poses, ranging from minor movement to touching or missing fingers.
 [63]

Gayane Shalunts, Yll Haxhimusa, and Robert Sablatnig.
Architectural style classification of building facade windows.
In George Bebis, Richard D. Boyle, Bahram Parvin, Darko Koracin, Song
Wang, Kyungnam Kim, Bedrich Benes, Kenneth Moreland, Christoph W. Borst,
Stephen DiVerdi, YiJen Chiang, and Jiang Ming, editors, Advances in
Visual Computing  7th International Symposium, ISVC 2011, Las Vegas, NV,
USA, September 2628, 2011. Proceedings, Part II, volume 6939 of
Lecture Notes in Computer Science, pages 280289. Springer, 2011.
[ http 
.pdf ]
Building facade classification by architectural styles allows categorization of large databases of building images into semantic categories belonging to certain historic periods, regions and cultural influences. Image databases sorted by architectural styles permit effective and fast image search for the purposes of contentbased image retrieval, 3D reconstruction, 3D citymodeling, virtual tourism and indexing of cultural heritage buildings. Building facade classification is viewed as a task of classifying separate architectural structural elements, like windows, domes, towers, columns, etc, as every architectural style applies certain rules and characteristic forms for the design and construction of the structural parts mentioned. In the context of building facade architectural style classification the current paper objective is to classify the architectural style of facade windows. Typical windows belonging to Romanesque, Gothic and Renaissance/Baroque E uropean main architectural periods are classified. The approach is based on clustering and learning of local features, applying intelligence that architects use to classify windows of the mentioned architectural styles in the training stage.
 [62]

Michael Gerstmayer, Yll Haxhimusa, and Walter G. Kropatsch.
Hierarchical interactive image segmentation using irregular pyramids.
In Xiaoyi Jiang, Miquel Ferrer, and Andrea Torsello, editors,
GraphBased Representations in Pattern Recognition  8th IAPRTC15
International Workshop, GbRPR 2011, Münster, Germany, May 1820, 2011.
Proceedings, volume 6658 of Lecture Notes in Computer Science, pages
245254. Springer, 2011.
[ http 
.pdf ]
In this paper we describe modifications of irregular image segmentation pyramids based on userinteraction. We first build a hierarchy of segmentations by the minimum spanning tree based method, then regions from different (granularity) levels are combined to a final (better) segmentation with userspecified operations guiding the segmentation process. Based on these operations the users can produce a final image segmentation that best suits their applications. This work can be used for applications where we need accuracy in image segmentation, in annotating images or creating ground truth among others.
 [61]

Yll Haxhimusa, Edward Carpenter, Joseph Catrambone, David Foldes, Emil
Stefanov, Laura Arns, and Zygmunt Pizlo.
2d and 3d traveling salesman problem.
Journal of Problem Solving, 2010.
[ http 
.pdf ]
When a twodimensional (2D) traveling salesman problem (TSP) is presented on a computer screen, human subjects can produce nearoptimal tours in linear time. In this study we tested human performance on a real and virtual floor, as well as in a threedimensional (3D) virtual space. Human performance on the real floor is as good as that on a computer screen. Performance on a virtual floor is very similar, while that in a 3D space is slightly but systematically worse. We modeled these results by a graph pyramid algorithm. The same algorithm can account for the results with 2D and 3D problems, which suggests that deterioration of performance in the 3D space can be attributed to geometrical relations between hierarchical clustering in a 3D space and coarsetofine production of a tour.
 [60]
 Fuensanta Torres, Rebeca Marfil, Yll Haxhimusa, and Antonio Bandera. Combining regular decimation and dual graph contraction for hierarchical image segmentation. In Rocio GonzalezDiaz and Pedro Real Jurado, editors, 3rd International Workshop on Computational Topology in Image Context, pages 97104, Chipiona  Spain, November 2010. University of Sevilla, Spain. ISSN: 18854508. [ .pdf ]
 [59]

Adrian Ion, Yll Haxhimusa, and Walter G. Kropatsch.
An improved coordinate system for point correspondences of 2d
articulated shapes.
In Srecko Brlek, Christophe Reutenauer, and Xavier Provençal,
editors, Discrete Geometry for Computer Imagery, 15th IAPR International
Conference, DGCI 2009, Montréal, Canada, September 30  October 2, 2009.
Proceedings, volume 5810 of Lecture Notes in Computer Science, pages
92103. Springer, 2009.
[ http 
.pdf ]
To find corresponding points in different poses of the same articulated shape, a non rigid coordinate system is used. Each pixel of each shape is identified by a pair of distinct coordinates. The coordinates are used to address corresponding points. This paper proposes a solution to a discretization problem identified in a previous approach. The polar like coordinate system is computed in a space where the problem cannot occur, followed by mapping the computed coordinates to pixels.
 [58]
 Martin Stubenschrott, Walter G. Kropatsch, and Yll Haxhimusa. Combining an optical flow feature detector with graphbased segmentation. In A. Ion and W.G. Kropatsch, editors, Proceedings of the 14th Computer Vision Winter Workshop, CVWW 2009, pages 9198, Eibiswald, Austria, February 2009. Pattern Recognition and Image Processing Lab, Vienna University of Technology. [ .pdf ]
 [57]
 Adrian Ion, Yll Haxhimusa, and Walter G. Kropatsch. A coordinate system for articulated 2d shape point correspondences (further aligning theta). In A. Ion and W.G. Kropatsch, editors, Proceedings of the 14th Computer Vision Winter Workshop, CVWW 2009, pages 9198, Eibiswald, Austria, February 2009. Pattern Recognition and Image Processing Lab, Vienna University of Technology. [ .pdf ]
 [56]

Yll Haxhimusa, Walter G. Kropatsch, Zygmunt Pizlo, and Adrian Ion.
Approximative graph pyramid solution of the etsp.
Image Vision Comput., 27(7):887896, 2009.
[ http ]
The traveling salesman problem (TSP) is difficult to solve for input instances with large number of cities. Instead of finding the solution for an input with a large number of cities, the problem is transformed into a simpler form containing smaller number of cities, which is then solved optimally. Graph pyramid solution strategies, using Boruvka's minimum spanning tree step, convert, in a bottomup processing, a 2D Euclidean TSP problem with a large number of cities into successively smaller problems (graphs) with similar layout and solution, until the number of cities is small enough to seek the optimal solution. Expanding this tour solution in a topdown manner, to the lower levels of the pyramid, leads to an approximate solution. The new model has an adaptive spatial structure and it simulates visual acuity and visual attention. The model solves the TSP problem sequentially, by moving attention from city to city, and the quality of the solutions is similar to the solutions produced by humans. The graph pyramid data structures and processing strategies provide good methods for finding nearoptimal solutions for computationally hard problems. Isolating processing used by humans to solve computationally hard problems is of general importance to psychology community and might lead to advances in pattern recognition.
 [55]

Samuel Peltier, Adrian Ion, Walter G. Kropatsch, Guillaume Damiand, and Yll
Haxhimusa.
Directly computing the generators of image homology using graph
pyramids.
Image Vision Comput., 27(7):846853, 2009.
[ http ]
We introduce a method for computing homology groups and their generators of a 2D image, using a hierarchical structure, i.e. irregular graph pyramid. Starting from an image, a hierarchy of the image is built by two operations that preserve homology of each region. Instead of computing homology generators in the base where the number of entities (cells) is large, we first reduce the number of cells by a graph pyramid. Then homology generators are computed efficiently on the top level of the pyramid, since the number of cells is small. A top down process is then used to deduce homology generators in any level of the pyramid, including the base level, i.e. the initial image. The produced generators fit on the object boundaries. A unique set of generators called the minimal set, is defined and its computation is discussed. We show that the new method produces valid homology generators and present some experimental results.
 [54]

Adrian Ion, Yll Haxhimusa, Walter G. Kropatsch, and Salvador B. López
Mármol.
A coordinate system for articulated 2d shape point correspondences.
In 19th International Conference on Pattern Recognition (ICPR
2008), December 811, 2008, Tampa, Florida, USA, pages 14. IEEE, 2008.
[ http 
.pdf ]
A framework for mapping a polarlike coordinate system to a nonrigid shape is presented. Using a graph pyramid, a binary shape is decomposed into connected parts, based on its structure as captured by the eccentricity transform. The decomposition is used to derive domains for the angular like coordinate. A closest point search is employed to find point correspondences.
 [53]

Yll Haxhimusa, Zygmunt Pizlo, and Joseph Catrambone.
Noneuclidean visual traveling salesman problem.
Journal of Vision, 8(6):941, 2008.
[ http ]
Traveling Salesman Problem (TSP) is defined as the task of finding the shortest tour of N cities given intercity costs. Usually the intercity costs are 2D Euclidean distances. In the presence of obstacles or in the case of 3D surfaces, the intercity distances are in general not Euclidean. The TSP with obstacles and on 3D surfaces approximates our everyday visual navigation. There are three questions related to the mechanisms involved in solving TSP: (i) how do subjects find the intercity distances, (ii) how do they determine clusters of cities, and (ii) how do they produce the TSP tour. In our model, the nonEuclidean distances (geodesics) are found by using a nonlinear Eikonal equation, i.e. the evolution of interfaces (Sethian, 1999). The geodesic distances are then used as intercity costs in an MST graph pyramid (Haxhimusa et. al., 2007). The original TSP problem is represented by a sequence of problems involving clusters of cities. The hierarchical clustering is performed by using a Boruvka's minimum spanning tree. Close to the top of the pyramid, the original TSP problem is represented at a very coarse level and involves very small number of 'cities'. This coarse representation is solved optimally. Expanding this coarse tour in a topdown manner leads to a solution of the original TSP. The new model has an adaptive spatial structure and it simulates visual acuity and visual attention. The model solves the TSP problem sequentially, by moving its attention from city to city. The model's performance will be compared to the performance of human subjects.
 [52]
 Yll Haxhimusa, Joseph Catrambone, and Zygmunt Pizlo. Traveling salesman problem with noneuclidean distances. Washington DC, USA, 2008. abstract.
 [51]

Axel Pinz, Horst Bischof, Walter G. Kropatsch, Gerald Schweighofer, Yll
Haxhimusa, Andreas Opelt, and Adrian Ion.
Representations for cognitive vision: A review of appearancebased,
spatiotemporal, and graphbased approaches.
Electronic Letters on Computer Vision and Image Analysis,
7(2):3561, 2008.
[ http 
.pdf ]
The emerging discipline of cognitive vision requires a proper representation of visual information including spatial and temporal relationships, scenes, events, semantics and context. The goal of this review article is to summarize existing representational schemes which might be useful for cognitive vision, and to discuss promising future research directions. We structure the various approaches into appearancebased, spatiotemporal and graphbased representations for cognitive vision. The representation of objects has been covered extensively in computer vision research, both from a reconstruction as well as from a recognition point of view. Cognitive vision, however, will also require new ideas how to represent scenes. We introduce new concepts for scene representations and discuss how these might be eciently implemented in future cognitive vision systems.
 [50]
 Yll Haxhimusa, Emil Stefanov, and Pizlo Zygmunt. Solving Euclidean Traveling Salesman Problems in the Presence of Errors. Irvine, California, July, 2007. Abstract.
 [49]

Andreas Hubmer, Adrian Ion, Walter G. Kropatsch, Yll Haxhimusa, and Hubert
Hausegger.
How humans describe short videos  details of an experiment.
Technical Report PRIPTR113, Pattern Recognition and Image
Processing Lab, Institute of Computer Aided Automation, Vienna University of
Technology, June 2007.
[ .pdf 
.pdf ]
Human vision can be used as a model for computer vision. We have conducted an experiment to investigate several properties of human vision that can be applied to, and that can improve computer vision. This report describes in details the description of videos done by human subjects. Human descriptions of videos show the importance of higher levels of abstraction and that features of an object related to a task can raise the object's relevance.
 [48]
 Yll Haxhimusa. Structurally Optimal Dual Graph Pyramid and its Application in Image Partitioning. Dissertations in Artificial Intelligence. IOS Press and Akadademische Verlagsgesellschaft AKA, Berlin, 2006. [ http ]
 [47]

Adrian Ion, Samuel Peltier, Yll Haxhimusa, and Walter G. Kropatsch.
Decomposition for efficient eccentricity transform of convex shapes.
In Walter G. Kropatsch, Martin Kampel, and Allan Hanbury, editors,
Computer Analysis of Images and Patterns, 12th International Conference,
CAIP 2007, Vienna, Austria, August 2729, 2007, Proceedings, volume 4673 of
Lecture Notes in Computer Science, pages 653660. Springer, 2007.
[ http 
.pdf ]
The eccentricity transform associates to each point of a shape the shortest distance to the point farthest away from it. It is defined in any dimension, for open and closed manyfolds. Topdown decomposition of the shape can be used to speed up the computation, with some partitions being better suited than others. We study basic convex shapes and their decomposition in the context of the continuous eccentricity transform. We show that these shapes can be decomposed for a more efficient computation. In particular, we provide a study regarding possible decompositions and their properties for the ellipse, the rectangle, and a class of elongated shapes.
 [46]
 Adrian Ion, Gabriel Peyré, Yll Haxhimusa, Samuel Peltier, Walter G. Kropatsch, and Laurent Cohen. Shape matching using the geodesic eccentricity transform  a study. In C. Beleznai W. Ponweiser, M. Vincze, editor, The 31st annual workshop of the Austrian Association for Pattern Recognition OAGM 2007), pages 97104, Schloss Krumbach, Austria, May 2007. Austrian Computer Society. [ .pdf ]
 [45]

Yll Haxhimusa, Walter G. Kropatsch, Zygmunt Pizlo, Adrian Ion, and Andreas
Lehrbaum.
Approximating tsp solution by mst based graph pyramid.
In Francisco Escolano and Mario Vento, editors, GraphBased
Representations in Pattern Recognition, 6th IAPRTC15 International
Workshop, GbRPR 2007, Alicante, Spain, June 1113, 2007, Proceedings, volume
4538 of Lecture Notes in Computer Science, pages 295306. Springer,
2007.
[ http 
.pdf ]
The traveling salesperson problem (TSP) is difficult to solve for input instances with large number of cities. Instead of finding the solution of an input with a large number of cities, the problem is approximated into a simpler form containing smaller number of cities, which is then solved optimally. Graph pyramid solution strategies, in a bottomup manner using Boruvka's minimum spanning tree, convert a 2D Euclidean TSP problem with a large number of cities into successively smaller problems (graphs) with similar layout and solution, until the number of cities is small enough to seek the optimal solution. Expanding this tour solution in a topdown manner to the lower levels of the pyramid approximates the solution. The new model has an adaptive spatial structure and it simulates visual acuity and visual attention. The model solves the TSP problem sequentially, by moving attention from city to city with the same quality as humans. Graph pyramid data structures and processing strategies are a plausible model for finding nearoptimal solutions for computationally hard pattern recognition problems.
 [44]

Samuel Peltier, Adrian Ion, Yll Haxhimusa, Walter G. Kropatsch, and Guillaume
Damiand.
Computing homology group generators of images using irregular graph
pyramids.
In Francisco Escolano and Mario Vento, editors, GraphBased
Representations in Pattern Recognition, 6th IAPRTC15 International
Workshop, GbRPR 2007, Alicante, Spain, June 1113, 2007, Proceedings, volume
4538 of Lecture Notes in Computer Science, pages 283294. Springer,
2007.
[ http 
.pdf ]
We introduce a method for computing homology groups and their generators of a 2D image, using a hierarchical structure i.e. irregular graph pyramid. Starting from an image, a hierarchy of the image is built, by two operations that preserve homology of each region. Instead of computing homology generators in the base where the number of entities (cells) is large, we first reduce the number of cells by a graph pyramid. Then homology generators are computed efficiently on the top level of the pyramid, since the number of cells is small, and a top down process is then used to deduce homology generators in any level of the pyramid, including the base level i.e. the initial image. We show that the new method produces valid homology generators and present some experimental results.
 [43]
 Sammuel Peltier, Adrian Ion, Yll Haxhimusa, and Walter G. Kropatsch. Controlling geometry of homology generators. In M. Grabner and H. Grabner, editors, Proceedings of the 12th Computer Vision Winter Workshop, CWVV 2007, pages 115121, St. Lambrecht, Austria, February 2007. Austrian Computer Society. [ .pdf ]
 [42]
 Walter G. Kropatsch, Yll Haxhimusa, and Adrian Ion. Multiresolution image segmentations in graph pyramids. In Abraham Kandel, Horst Bunke, and Mark Last, editors, Applied Graph Theory in Computer Vision and Pattern Recognition, volume 52 of Studies in Computational Intelligence, pages 341. Springer, 2007. [ http ]
 [41]
 Walter G. Kropatsch, Yll Haxhimusa, and Pizlo Zygmunt. MST based Pyramid Model of TSP. Vancouver, Canada, August, 2006. Abstract. [ .pdf ]
 [40]

Samuel Peltier, Adrian Ion, Yll Haxhimusa, and Walter G. Kropatsch.
Computing homology group generators of images using irregular graph
pyramids.
Technical Report PRIPTR111, Pattern Recognition and Image
Processing Lab, Institute of Computer Aided Automation, Vienna University of
Technology, December 2006.
[ .pdf 
.pdf ]
We introduce a method for computing homology groups and their generators of a 2D image, using a hierarchical structure i.e. an irregular graph pyramid. Instead of computing homology generators in the base where the number of entities (cells) is large, we first reduce the number of cells by a graph pyramid. Then homology generators are computed efficiently on the top level of the pyramid, since the number of cells is small, and a top down process is then used to deduce homology generators in any level of the pyramid, including the base level i.e. the initial image. We show that the new method produces valid homology generators and present some experimental results. In this report we also show that the generators of the first homology groups of a 2D image, computed with this pyramid based method always fit on the borders of the regions.
 [39]

Axel Pinz, Horst Bischof, Walter Kropatsch, Gerald Schweighofer, Yll Haxhimusa,
Andreas Opelt, and Adrian Ion.
Representations for cognitive vision: A review of appearancebased,
spatiotemporal, and graphbased approaches.
Technical Report PRIPTR109 and EMT200602, Pattern Recognition and
Image Processing Lab, Institute of Computer Aided Automation, Vienna
University of Technology, 2006.
[ .pdf 
.pdf ]
The emerging discipline of cognitive vision requires a proper representation of visual information including spatial and temporal relationships, scenes, events, semantics and context. The goal of this review article is to summarize existing representational schemes which might be useful for cognitive vision, and to discuss promising future research directions. We structure the various approaches into appearancebased, spatiotemporal and graphbased representations for cognitive vision. The representation of objects has been covered extensively in computer vision research, both from a reconstruction as well as from a recognition point of view. Cognitive vision, however, will also require new ideas how to represent scenes. We introduce new concepts for scene representations and discuss how these might be efficiently implemented in future cognitive vision systems.
 [38]

Thomas Illetschko, Adrian Ion, Yll Haxhimusa, and Walter G. Kropatsch.
Effective programming of combinatorial maps using coma  a c++
framework for combinatorial maps.
Technical Report PRIPTR106, Pattern Recognition and Image
Processing Lab, Institute of Computer Aided Automation, Vienna University of
Technology, 2006.
[ .pdf 
.pdf ]
Combinatorial maps and pyramids have been studied in great detail in the past, and it has been shown that this concept is advantageous for many applications in the field of image processing and pattern recognition by providing means to store information of the topological relations of the represented data. In the course of these studies, the properties of combinatorial maps have been investigated using different sets of permutations, different operations and different algorithms. In each case new software had to be created in order to conduct experiments, as the existing programs were designed to work only for a specific model. Due to the complexity of combinatorial maps, the implementation of such a software is a time and resource intensive task. Thus these programming efforts were often responsible for delaying the presentation of new results in the past. This paper presents COMA  a C++ framework for combinatorial maps  that has been created during recent studies of combinatorial maps, motivated by this problem. Using an object oriented approach, COMA was specifically designed to allow an efficient and quick integration of changes to the model of combinatorial maps used, as well as the implementation of new algorithms. As a consequence COMA significantly reduces the amount of time needed to set up new experiments.
 [37]

Zygmunt Pizlo, Emil Stefanov, John Saalweachter, Zheng Li, Yll
Haxhimusa, and Walter G. Kropatsch.
Traveling salesman problem: a foveating model.
Journal of Problem Solving, 1(1):83101, 2006.
[ http 
.pdf ]
We tested human performance on the Euclidean Traveling Salesman Problem using problems with 650 cities. Results confirmed our earlier findings that: (a) the time of solving a problem is proportional to the number of cities, and (b) the solution error grows very slowly with the number of cities. We formulated a new version of a pyramid model. The new model has an adaptive spatial structure, and it simulates visual acuity and visual attention. Specifically, the model solves the ETSP problem sequentially by moving attention from city to city, the same way human subjects do. The model includes a parameter representing the magnitude of local search. This parameter allows modeling individual differences among the subjects. The computational complexity of the current implementation of the model is O(n2), but this can most likely be improved to O[nlog(n)]. Simulation experiments demonstrated psychological plausibility of the new model.
 [36]

Yll Haxhimusa, Adrian Ion, and Walter G. Kropatsch.
Irregular pyramid segmentations with stochastic graph decimation
strategies.
In José Francisco Martínez Trinidad, Jesús Ariel
CarrascoOchoa, and Josef Kittler, editors, Progress in Pattern
Recognition, Image Analysis and Applications, 11th Iberoamerican Congress in
Pattern Recognition, CIARP 2006, Cancun, Mexico, November 1417, 2006,
Proceedings, volume 4225 of Lecture Notes in Computer Science, pages
277286. Springer, 2006.
[ http 
.pdf ]
In this paper we use different decimation strategies in irregular pyramid segmentation framework, to produce perceptually important groupings. These graph decimation strategies, based on the maximum independent set concept, used in Boruvka's minimum spanning tree based partitioning method, show similar discrepancy segmentation errors. Global and local consistency error measures do not show big differences between the methods although human visual inspection of the results show advantages for one method. To a certain extent this subjective impression is captured by the new criteria of 'region size variation'.
 [35]

Adrian Ion, Thomas Illetschko, Yll Haxhimusa, and Walter G. Kropatsch.
Distinguishing 3dtopological configurations of two tori.
In Viorel Negru, Dana Petcu, Daniela Zaharie, Ajith Abraham, Bruno
Buchberger, Alexandru Cicortas, Dorian Gorgan, and Joël Quinqueton,
editors, 8th International Symposium on Symbolic and Numeric Algorithms
for Scientific Computing (SYNASC 2006), 2629 September 2006, Timisoara,
Romania, pages 111118. IEEE Computer Society, 2006.
[ http 
.pdf ]
Most of the existing work regarding topology preserving hierarchies is mainly preoccupied with 2D domains. But recently attention has turned to 3D, and more generally, nD representations. Even more than in 2D, the necessity for reducing these representations exists and motivates the research in hierarchical structures i.e. pyramids. Using representations that support any dimension, like e.g. the combinatorial map, n dimensional irregular pyramids can be built, thus obtaining reduced representations of the original data, while preserving the topology. This paper presents 3D combinatorial maps and the primitive operations needed to simplify such representations. Minimal configurations of the three primitive topological configurations, simplex, hole, and tunnel, and two possible configurations for two tori are presented. Experimental results and possible applications show the potential of the approach.
 [34]

Walter G. Kropatsch, Adrian Ion, Yll Haxhimusa, and Thomas Flanitzer.
The eccentricity transform (of a digital shape).
In Attila Kuba, László G. Nyúl, and Kálmán
Palágyi, editors, Discrete Geometry for Computer Imagery, 13th
International Conference, DGCI 2006, Szeged, Hungary, October 2527, 2006,
Proceedings, volume 4245 of Lecture Notes in Computer Science, pages
437448. Springer, 2006.
[ http 
.pdf ]
Eccentricity measures the shortest length of the paths from a given vertex v to reach any other vertex w of a connected graph. Computed for every vertex v it transforms the connectivity structure of the graph into a set of values. For a connected region of a digital image it is defined through its neighbourhood graph and the given metric. This transform assigns to each element of a region a value that depends on it's location inside the region and the region's shape. The definition and several properties are given. Presented experimental results verify its robustness against noise, and its increased stability compared to the distance transform. Future work will include using it for shape decomposition, representation, and matching.
 [33]

Yll Haxhimusa, Adrian Ion, and Walter G. Kropatsch.
Evaluating hierarchical graphbased segmentation.
In 18th International Conference on Pattern Recognition (ICPR
2006), Volume 2, 2024 August 2006, Hong Kong, China, pages 195198. IEEE
Computer Society, 2006.
[ http 
.pdf ]
Using real world images, two hierarchical graphbased segmentation methods are evaluated with respect to segmentations produced by humans. Global and local consistency measures do not show big differences between the two representative methods although human visual inspection of the results show advantages for one method. To a certain extent this subjective impression is captured by the new criteria of 'region size variation.
 [32]

Yll Haxhimusa.
Structurally Optimal Dual Graph Pyramid and its Application in
Image Partitioning.
PhD thesis, Vienna University of Technology, Faculty of Informatics,
Institute of Computer Aided Automation, Pattern Recognition and Image
Processing Group, 2006.
[ .pdf ]
A widely used hierarchical representation in many areas of computer vision and pattern recognition is the (regular) image pyramid, which employs both coarse to fine and fine to coarse processing strategies. Regular pyramids rapidly compute global information in a recursive manner, because their height is logarithmically bounded by the size of the input. Regular image pyramids lack shift invariance as a result of the fixed interlevel neighborhood. Irregular hierarchical structures (irregular pyramids) overcome shift invariance, among others. However, their logarithmic height cannot be guaranteed in general, as well as the computational efficiency. Main topics of this work are irregular graph pyramids and their application in image partitioning. We introduce two new decimation concepts, maximal independent edge set (MIES) and maximal independent directed edge set (MIDES), both based on the maximal independent set principle. We show that the construction of stochastic irregular pyramids bounds logarithmically the height of the pyramid. Within this irregular graph pyramid framework, we introduce a time efficient image partitioning method based on the minimum spanning tree principle.
 [31]

Adrian Ion, Walter G. Kropatsch, and Yll Haxhimusa.
Considerations regarding the minimum spanning tree pyramid
segmentation method.
In DitYan Yeung, James T. Kwok, Ana L. N. Fred, Fabio Roli, and Dick
de Ridder, editors, Structural, Syntactic, and Statistical Pattern
Recognition, Joint IAPR International Workshops, SSPR 2006 and SPR 2006, Hong
Kong, China, August 1719, 2006, Proceedings, volume 4109 of Lecture
Notes in Computer Science, pages 182190. Springer, 2006.
[ http 
.pdf ]
The minimum spanning tree pyramid is a hierarchical image segmentation method. We study it's properties and the regions it produces. We show the similarity with the watershed transform and present the method in a domain in which this is easy to understand. For this, a short overview of both methods is given. Catchment basins are contracted before their neighbouring local maximas. Smooth regions surrounded by borders with maximal local variation are selected. The maximum respectively minimum variation on the border of a region is larger than the maximum respectively minimum variation inside the region.
 [30]
 Adrian Ion, Hubert Hausegger, Walter G. Kropatsch, and Yll Haxhimusa. How humans describe short videos. In Markus Vincze and Lucas Paletta, editors, Proceedings of the 2nd International Cognitive Vision Workshop, Graz, Austria, 13, May 2006. [ .pdf ]
 [29]
 Yll Haxhimusa, Adrian Ion, and Walter G. Kropatsch. Comparing hierarchies of segmentations: Humans, normalized cut, and minimum spanning tree. In Frank Lenzen, Otmar Scherzer, and Markus Vincze, editors, Proceedings of 30th OEAGM Workshop, pages 95103, Obergurgl, Austria, 2006. Austrian Computer Society. [ .pdf ]
 [28]
 Thomas Illetschko, Adrian Ion, Yll Haxhimusa, and Walter G. Kropatsch. Collapsing 3d combinatorial maps. In Frank Lenzen, Otmar Scherzer, and Markus Vincze, editors, Proceedings of 30th Austrian Association of Pattern Recognition Workshop, pages 8593, Obergurgl, Austria, 2006. Austrian Computer Society. [ .pdf ]
 [27]
 Thomas Illetschko, Adrian Ion, Yll Haxhimusa, and Walter G. Kropatsch. Distinguishing the 3 primitive 3dtopological configurations: Simplex, hole, tunnel. In Ondrej Cum and Vojtech Franc, editors, Proceedings of the 11th. Computer Vision Winter Workshop CVWW 2006, pages 2227, Telc, Czech Republic, February 2006. Czeck Pattern Recognition Society. [ .pdf ]
 [26]

Walter G. Kropatsch, Yll Haxhimusa, and Pascal Lienhardt.
Hierarchies relating topology and geometry.
In Henrik I. Christensen and HansHellmut Nagel, editors,
Cognitive Vision Systems, Sampling the Spectrum of Approaches [based on a
Dagstuhl seminar], volume 3948 of Lecture Notes in Computer Science,
pages 199220. Springer, 2006.
[ http 
.pdf ]
Cognitive Vision has to represent, reason and learn about objects in its environment it has to manipulate and react to. There are deformable objects like humans which cannot be described easily in simple geometric terms. In many cases they are composed of several pieces forming a 'structured subset' of R^n or Z^n . We introduce the potential topological representations for structured objects: plane graphs, combinatorial and generalized maps. They capture abstract spatial relations derived from geometry and enable reconstructions through attributing the relations by e.g. coordinates. In addition they offer the possibility to combine both topology and geometry in a hierarchical framework: irregular pyramids. The basic operations to construct these hierarchies are edge contraction and edge removal. We show preliminary results in using them to hold a whole set of segmentations of an image that enable reasoning and planning actions at various levels of detail down to a single pixel in a homogeneous way. We further speculate that the higher levels map the inherent structure of objects and can be used to integrate (and 'learn') the specific object properties over time by upprojecting individual measurements. The construction of the hierarchies follows the philosophy to reduce the data amount at each higher level of the hierarchy by a reduction factor > 1 while preserving important topological properties like connectivity and inclusion.
 [25]

Adrian Ion, Yll Haxhimusa, and Walter G. Kropatsch.
A graphbased concept for spatiotemporal information in cognitive
vision.
Technical Report PRIPTR098, Pattern Recognition and Image
Processing Lab, Institute of Computer Aided Automation, Vienna University of
Technology, 2005.
[ .pdf 
.pdf ]
A concept relating storyboard description of video sequences with spatiotemporal hierarchies build by local contraction processes of spatiotemporal relations is presented. Object trajectories are curves in which their ends and junctions are identified. Junction points happen when two (or more) trajectories touch or cross each other, which we interpret as the 'interaction' of two objects. Trajectory connections are interpreted as the high level descriptions.
 [24]
 Adrian Ion, Yll Haxhimusa, and Walter G. Kropatsch Kropatsch. A TopologyBased Concept for Contraction in Spatiotemporal Space. In M. Zillich and M. Vinze, editors, Proceedings of the 1st Austrian Cognitive Vision Workshop, pages 1926. Austrian Computer Society, February 2005. [ .pdf ]
 [23]
 Adrian Ion, Yll Haxhimusa, Walter G. Kropatsch, and Luc Brun. Hierarchical Image Partitioning using Combinatorial Maps. In A. Hanbury and H. Bischof, editors, Proceedings of the 10th Computer Vision Winter Workshop , CVWW 2005, pages 4352, Zell an der Pram, Austria, February 2005. Austrian Computer Society. [ .pdf ]
 [22]

Adrian Ion, Yll Haxhimusa, and Walter G. Kropatsch.
A graphbased concept for spatiotemporal information in cognitive
vision.
In Luc Brun and Mario Vento, editors, GraphBased
Representations in Pattern Recognition, 5th IAPR InternationalWorkshop, GbRPR
2005, Poitiers, France, April 1113, 2005, Proceedings, volume 3434 of
Lecture Notes in Computer Science, pages 223232. Springer, 2005.
[ http 
.pdf ]
A concept relating storyboard description of video sequences with spatiotemporal hierarchies build by local contraction processes of spatiotemporal relations is presented. Object trajectories are curves in which their ends and junctions are identified. Junction points happen when two (or more) trajectories touch or cross each other, which we interpret as the 'interaction' of two objects. Trajectory connections are interpreted as the high level descriptions.
 [21]

Yll Haxhimusa, Adrian Ion, Walter G. Kropatsch, and Thomas Illetschko.
Evaluating minimum spanning tree based segmentation algorithms.
In André Gagalowicz and Wilfried Philips, editors, Computer
Analysis of Images and Patterns, 11th International Conference, CAIP 2005,
Versailles, France, September 58, 2005, Proceedings, volume 3691 of
Lecture Notes in Computer Science, pages 579586. Springer, 2005.
[ http 
.pdf ]
Two segmentation methods based on the minimum spanning tree principle are evaluated with respect to each other. The hierarchical minimum spanning tree method is also evaluated with respect to human segmentations. Discrepancy measure is used as best suited to compute the segmentation error between the methods. The evaluation is done using gray value images. It is shown that the segmentation results of these methods have a considerable difference.
 [20]

Walter G. Kropatsch, Yll Haxhimusa, Zygmunt Pizlo, and Georg Langs.
Vision pyramids that do not grow too high.
Pattern Recognition Letters, 26(3):319337, 2005.
[ http 
.pdf ]
In irregular pyramids, their vertical structure is not determined beforehand as in regular pyramids. We present three methods, all based on maximal independent sets from graph theory, with the aim to simulate the major sampling properties of the regular counterparts: good coverage of the higher resolution level, not too large sampling gaps and, most importantly, the resulting height, e.g. the number of levels to reach the apex. We show both theoretically and experimentally that the number of vertices can be reduced by a factor of 2.0 at each level. The plausibility of log (diameter) pyramids is supported by psychological and psychophysical considerations. Their technical relevance is demonstrated by enhancing appearancebased object recognition. An irregular pyramid hypothesis generation for robust PCA through topdown attention mechanisms achieves higher speed and quality than regular pyramids and nonpyramidal approaches.
 [19]

Walter G. Kropatsch and Yll Haxhimusa.
Grouping of nonconnected structures by an irregular graph pyramid.
In Jorge S. Marques, Nicolas Pérez de la Blanca, and Pedro Pina,
editors, Pattern Recognition and Image Analysis, Second Iberian
Conference, IbPRIA 2005, Estoril, Portugal, June 79, 2005, Proceedings, Part
II, volume 3523 of Lecture Notes in Computer Science, pages 107114.
Springer, 2005.
[ http 
.pdf ]
Motivated by claims to 'bridge' the representational gap between image and model features' and by the growing importance of topological properties we discuss several extensions to dual graph pyramids: structural simplification should preserve important topological properties and content abstraction could be guided by an external knowledge base. We review multilevel graph hierarchies under the special aspect of their potential for abstraction and grouping.
 [18]
 Yll Haxhimusa, Adrian Ion, Walter G. Kropatsch, and Luc Brun. Hierarchical Image Partitioning using Combinatorial Maps. In D. Chetverikov, L. Czuni, and M. Vincze, editors, Proceeding of the Joint HungarianAustian Conference on Image Processing and Pattern Recognition, HACIPPR 2005  OAGM 2005/KEPAF 2005, pages 179186, Veszprém, Hungary, 1113, May 2005. Austrian Computer Society. [ .pdf ]
 [17]

Yll Haxhimusa and Walter G. Kropatsch.
Segmentation graph hierarchies.
In Ana L. N. Fred, Terry Caelli, Robert P. W. Duin, Aurélio C.
Campilho, and Dick de Ridder, editors, Structural, Syntactic, and
Statistical Pattern Recognition, Joint IAPR International Workshops, SSPR
2004 and SPR 2004, Lisbon, Portugal, August 1820, 2004 Proceedings, volume
3138 of Lecture Notes in Computer Science, pages 343351. Springer,
2004.
[ http 
.pdf ]
The region's internal properties (color, texture, ...) help to identify them and their external relations (adjacency, inclusion, ...) are used to build groups of regions having a particular consistent meaning in a more abstract context. Lowlevel cue image segmentation in a bottomup way, cannot and should not produce a complete final 'good' segmentation. We present a hierarchical partitioning of images using a pairwise similarity function on a graphbased representation of an image. The aim of this paper is to build a minimum weight spanning tree (MST) of an image in order to find region borders quickly in a bottomup 'stimulusdriven' way based on local differences in a specific feature.
 [16]

Walter G. Kropatsch, Yll Haxhimusa, and Zygmunt Pizlo.
Integral trees: Subtree depth and diameter.
Technical Report PRIPTR092, Pattern Recognition and Image
Processing Lab, Institute of Computer Aided Automation, Vienna University of
Technology, 2004.
[ .pdf 
.pdf ]
Regions in an image graph can be described by their spanning tree. A graph pyramid is a stack of image graphs at different granularities. Integral features capture important properties of these regions and the associated trees. We compute the depth of a rooted tree, its diameter and the center which becomes the root in the topdown decomposition of a region. The integral tree is an intermediate representation labeling each vertex of the tree with the integral feature(s) of the subtree. Parallel algorithms efficiently compute the integral trees for subtree depth and diameter enabling local decisions with global validity in subsequent topdown processes.
 [15]
 Walter G. Kropatsch and Yll Haxhimusa. Grouping and Segmentation in a Hierarchy of Graphs. In Charlie A. Bouman and Eric L. Miller, editors, Proceedings of 16th Annual IS&T/SPIE Symposium Electronic Imaging, Computational Imaging II, volume SPIE 5299, pages 193204, USA, 2004. IS&T/SPIE. [ .pdf ]
 [14]
 Walter G. Kropatsch and Yll Haxhimusa. Hierarchical Grouping of Nonconnected Structures. In Wilhelm Burger and Josef Scharinger, editors, Proceedings of 28th Austrian Association of Pattern Recognition Workshop, OEAGM 2004, pages 165172, Hagenberg, Austria, 2004. OCG. [ .pdf ]
 [13]

Walter G. Kropatsch, Yll Haxhimusa, and Zygmunt Pizlo.
Integral trees: Subtree depth and diameter.
In Reinhard Klette and Jovisa D. Zunic, editors, Combinatorial
Image Analysis, 10th InternationalWorkshop, IWCIA 2004, Auckland, New
Zealand, December 13, 2004, Proceedings, volume 3322 of Lecture Notes
in Computer Science, pages 7787. Springer, 2004.
[ http 
.pdf ]
 [12]

Yll Haxhimusa, Roland Glantz, and Walter G. Kropatsch.
Constructing stochastic pyramids by mides  maximal independent
directed edge set.
In Edwin R. Hancock and Mario Vento, editors, Graph Based
Representations in Pattern Recognition, 4th IAPR International Workshop,
GbRPR 2003, York, UK, June 30  July 2, 2003, Proceedings, volume 2726 of
Lecture Notes in Computer Science, pages 2434. Springer, 2003.
[ http 
.pdf ]
We present a new method (MIDES) to determine contraction kernels for the construction of graph pyramids. Experimentally the new method has a reduction factor higher than 2.0. Thus, the new method yields a higher reduction factor than the stochastic decimation algorithm (MIS) and maximal independent edge set (MIES), in all tests. This means the number of vertices in the subgraph induced by any set of contractible edges is reduced to half or less by a single parallel contraction. The lower bound of the reduction factor becomes crucial with large images.
 [11]
 Yll Haxhimusa and Walter G. Kropatsch. Image Partitioning with Graph Pyramids. In C.Beleznai and Th. Schoegl, editors, Proceedings of 27th Austrian Association of Pattern Recognition Workshop, OEAGM 2003, pages 8188, Laxenburg, Austria, 2003. Austrian Computer Society. [ .pdf ]
 [10]

Yll Haxhimusa and Walter G. Kropatsch.
Hierarchy of partitions with dual graph contraction.
In Bernd Michaelis and Gerald Krell, editors, Pattern
Recognition, 25th DAGM Symposium, Magdeburg, Germany, September 1012, 2003,
Proceedings, volume 2781 of Lecture Notes in Computer Science, pages
338345. Springer, 2003.
[ http 
.pdf ]
We present a hierarchical partitioning of images using a pairwise similarity function on a graphbased representation of an image. This function measures the difference along the boundary of two components relative to a measure of differences of component's internal differences. This definition attempts to encapsulate the intuitive notion of contrast. Two components are merged if there is a lowcost connection between them. Each component's internal difference is represented by the maximum edge weight of its minimum spanning tree. External differences are the cheapest weight of edges connecting components. We use this idea to find region borders quickly and effortlessly in a bottomup 'stimulusdriven' way based on local differences in a specific feature, like as in preattentive vision. The components are merged ignoring the details in regions of highvariability, and preserving the details in lowvariability ones.
 [9]

Yll Haxhimusa and Walter G. Kropatsch.
Hierarchical Image Partitioning with Dual Graph Contraction.
Technical Report PRIPTR81, Pattern Recognition and Image Processing
Lab, Institute of Computer Aided Automation, Vienna University of Technology,
July 2003.
[ .pdf 
.pdf ]
We present a hierarchical partitioning of images using a pairwise similarity function on a graphbased representation of an image. This function measures the difference along the boundary of two components relative to a measure of differences of the components' internal differences. This definition tries to encapsulate the intuitive notion of contrast. Two components are merged if there is a lowcost connection between them. Each component's internal difference is represented by the maximum edge weight of its minimum spanning tree. External differences are the smallest weight of edges connecting components. We use this idea for building a minimum spanning tree to find region borders quickly and effortlessly in a bottomup way, based on local differences in a specific feature.
 [8]

Yll Haxhimusa and Walter G. Kropatsch.
Experimental results of mis, mies, mides and d3p.
Technical Report PRIPTR78, Pattern Recognition and Image Processing
Lab, Institute of Computer Aided Automation, Vienna University of Technology,
2003.
[ .pdf 
.pdf ]
In this technical report we present in detail the results of the first 100 experiments of stochastic irregular graph pyramid of 100x100 and 200x200 images i.e graphs using methods MIS, MIES, MIDES and D3P. For details about these methods and irregular images pyramid see Technical Report PRIPTR74. This report extends PRIPTR74.
 [7]
 Yll Haxhimusa, Roland Glantz, Maamar Saib, G. Langs, and Walter G. Kropatsch. Reduction Factors of Image Pyramid on Udirected and Directed Graph. In H. Wildenauer and W. Kropatsch, editors, Proceedings of the 7th. Computer Vision Winter Workshop, CVWW 2002, pages 2938. Pattern Recognition and Image Processing Lab, Institute of Computer Aided Automation, Vienna University of Technology, 2002. [ .pdf ]
 [6]

Yll Haxhimusa and Walter G. Kropatsch.
Reduction Factors of Pyramids on Undirected and Directed
Graphs.
Technical Report PRIPTR74, Pattern Recognition and Image Processing
Lab, Institute of Computer Aided Automation, Vienna University of Technology,
2002.
[ .pdf 
.pdf ]
We present two new methods to determine contraction kernels for the construction of graph pyramids. The first method is restricted to undirected graphs and yields a reduction factor of at least 2.0. This means with our method the number of vertices in the subgraph induced by any set of contractible edges reduces to half or less by a single parallel contraction. Our second method also works for directed graphs. Our methods yield better reduction factors than Meer's stochastic decimation algorithm, in all tests. The lower bound of the reduction factor becomes crucial with large images. We also give a method to compare the structure of the image pyramid.
 [5]
 Yll Haxhimusa and Walter G. Kropatsch. Path Lengths in Sochastic Graph Image Pyramid. In F. Leberl and F. Fraundorfer, editors, Proceedings of 26th OEAGM Workshop, pages 7986, Graz, 2002. OCG. [ .pdf ]
 [4]

Yll Haxhimusa, Roland Glantz, Maamar Saib, Georg Langs, and Walter G.
Kropatsch.
Logarithmic tapering graph pyramid.
In Luc J. Van Gool, editor, Pattern Recognition, 24th DAGM
Symposium, Zurich, Switzerland, September 1618, 2002, Proceedings, volume
2449 of Lecture Notes in Computer Science, pages 117124. Springer,
2002.
[ http 
.pdf ]
We present a new method to determine contraction kernels for the construction of graph pyramids. The new method works with undirected graphs and yields a reduction factor of at least 2.0. This means that with our method the number of vertices in the subgraph induced by any set of contractible edges is reduced to half or less by a single parallel contraction. Our method yields better reduction factors than the stochastic decimation algorithm, in all tests. The lower bound of the reduction factor becomes crucial with large images.
 [3]

Maamar Saib, Yll Haxhimusa, and Roland Glantz.
Dgc_tool: Building irregular graph pyramid using dual graph
contraction.
Technical Report PRIPTR69, Pattern Recognition and Image Processing
Lab, Institute of Computer Aided Automation, Vienna University of Technology,
2002.
[ .pdf 
.pdf ]
In this technical report the new version of the software Dgc tool is presented. This tool allows us to build up irregular graph pyramids by dual graph contraction. The graph pyramid consists of a stack of levels (pair of graphs), each of which has a primal level and its dual. Every successive level is a reduced version of the level below. Primal level and its dual represent a primal graph and its dual, respectively. The primal graph base level of the pyramid may represent a two dimensional image
 [2]
 Yll Haxhimusa, Maamar Saib, Roland Glantz, and Walter G. Kropatsch. Equivalent Contraction Kernels Using Dynamic Trees. In Boštjan Likar, editor, Computer Vision  CVWW'01, Proceedings of the 6th. Computer Vision Winter Workshop, pages 267275, Ljubljana, 2001. Slovenian Pattern Recognition Society.
 [1]
 Yll Haxhimusa. Rrjetat neurale për njohjen e numrave. Master's thesis, University of Prishtina, Faculty of Electrical Engineering, Department of Control and Electronics, May 1998, Prishtinë, Kosovë (in Albanian), 1998.
This file was generated by bibtex2html 1.95.