Sven Utcke's Research

Research Interests

Projective Geometry
Projective Geometry describes the relationship between the real, 3D, world and it's projection onto film or, these days, a CCD-chip. I've dabbled some with projective invariants4,6, that is measurements which remain unchanged under all possible projective transformations (and allow you to recognise an object from any image, no matter what it looks like in the image), but mainly I'm interested in the behaviour of (measurement) errors under projection6,7.
3D Reconstructions
Of course much of the motivation to study projective geometry stems from the desire to reconstruct the 3D-world from (one or more) 2D images. In the distant past I've mostly been concerned with tomographic methods for 3D-reconstruction1,2,3, but these days I am really more interested in methods which can work with a single image4,9,10.
Archaeometry is concerned with measurements on archaeological finds. I more or less happened on this by chance, since I had done some work on the reconstruction of rotationally symmetric objects from single images4, and since most archaeological finds are pottery, which is rotationally symmetric. However, here I'm not so much interested in the 3D reconstruction (although this is an interesting subject too, in particular when dealing with individual shards), but in the classification of objects based on their 2D-contour9.


Deutsch: Falls Sie ein Student sind und nach einer Studien oder Diplomarbeit ausschau halten, finden Sie hier(under construction) genauere Informationen. Insbesondere kann ich Arbeiten auf den Gebieten 3D-Rekonstruktion und Archäometry anbieten. Falls das Ihren Interessen nicht so entgegen kommt, sollten Sie auch mal einen Blick auf die anderen bei KOGS angebotenen Themen werfen.

English: If you are a student looking for a Diploma-thesis (Diplomarbeit), please note that I can offer several interesting subjects in particular in connection with 3D-reconstruction and archaeometry - some are described here(under construction), but you can always contact me for related topics. You might also want to have a look at other subjects currently on offer at KOGS.


3D Reconstruction

This is an ongoing research interest (since 1998) more than a real project. The idea here is to generate a qualitative depth-map from single images (which is theoretically impossible, but none the less constantly done by humans looking at pictures). In order to achive this goal I am utilising a-priori knowledge about the world (corners are usually right-angle, the floor is usually flat and horizontal, walls are usually vertical...) and combine it with some simple reasoning (T-junctions often mean that the part "above" the T is obscuring the part "below" the "T-bar",...), but of course also "proper" projective geometry. Unfortunately I have little time to work on this myself, but it would make a perfect Diplomarbeit. (under construction)


Archaeometry is the application of methods taken from the natural sciences - and in my case in particular computer vision - to archaeological problems. My interest in this area dates back to 1996, when Dr. Peter Werner and Prof. Machule from the TU Hamburg-Harburg asked Prof. Burkhard for help with the automatic classification of ancient ceramics, and he in turn asked me, due to my experience with rotationally symmetric objects4. Unfortunately this project too is mainly a spare-time activity and relies on Studien- and Diplomarbeiten. (under construction)


From 5/2001 till 6/2004 KOGS was a member of CogVis - Cognitive Vision Systems - a project sponsored by the EU. The aim of CogVis, as well as a number of similar projects also funded by the EU, was to "remarry" computer vision on the one hand and artificial intelligence on the other. Since KOGS is traditionally active in both fields we were a natural partner. The thumbnail shows part of an analysis of a table-laying scene, which is the scenario we concentrated on. Within CogVis I only acted as an advisor on the computer-vision aspects.
Find out more...

Grouping and Error-Propagation in Projective Geometry

This is the title of my yet to be viva-ed - never mind published - PhD-thesis. I mostly worked on this from 1995 till about 1999, with some added effort in 2003/2004. My interest in error-propagation started when, during the MOVIS-project6, I noticed that the application of straightforward projective geometry to my particular task (finding zebra-crossings) didn't work at all as expected - the measurements I had to work with where much too bad to be of any use. However, the error I observed was not well modelled by your usually Gaussians, and this set me thinking. The icon shows an error-model for the cross-ratio of four lines, an important projective invariant.
Find out more... (under construction)


From 1994 till 1997 I worked on the project MOVIS, the "Mobil Optoelectronic Visual Interpretative System for the blind and visually impaired", which would put two miniature cameras into normal glasses, connecting them to a computer and act as some sort of portable electronic guide dog, but one which can talk and read. Interestingly both KOGS (my current home) as well as Ti-1 (my former employer, and the one for which I worked during MOVIS) were part of MOVIS. Within MOVIS I was mostly concentrating on the recognition of street-signs; in particular pedestrian crossings but also other signs of interest, such as bus-stops, signs denoting public phones (not that many mobiles around in 1994) and such.
Find out more...

Defect Detection on Cigarette Packaging

In the autumn of 1994 we at the TU Hamburg-Harburg's Ti-1 were asked to conduct a study to assess the possibilities of automatic visual inspection of rather small defects on the cardboard printout for cigarette boxes. These are printed to surprisingly exacting standards (better then your average art print, I wouldn't be surprised), and even sub-millimetre defects are unwanted. The study is of course proprietary, so let me just say that yes, it could be done, even using 1994 technology and considering the tremendous speed at which theses things are printed (albeit just barely, and requiring careful lighting).

Transfer and Invariants of Surfaces of Revolution

From 7/1993 till 1/1994 I was staying with Dr. Zisserman in Oxford, working on my Diplomarbeit (approx. a MSc). Prof. Burkhard afforded me that unique opportunity, which combined intense and fruitful scientific work with a lot of fun! I rarely enjoyed any extended period of time as much as I enjoyed my stay in Oxford, and would like to thank all who made it possible. There I worked on the recognition (and reconstruction) of rotationally symmetric surfaces from single, projective views.
Find out more... (under construction)

Affine Alignment of a Brain (MRI) Atlas

From 4/1993 till 5/1993 I had some time on my hand in between the internship and some examinations on the one side and the start of my Diplomarbeit in Oxford on the other side, so I took up an offer of the Philips research laboratory to work on software which would map a recently acquired 3D-map of a human brain onto the actual MRI-scan of a real brain. Affinely map it, that is, by manually specifying 4 control points - not exactly the state of the art by todays standards, but in 1993 we were quite happy to get something like this displayed in reasonable time (3D-acceleration? Hell, you were lucky if your workstation supported 256 colours!). The main problem was to intelligently interpolate between individual slices once the atlas became rotated around the x- or y-axis.

Simulation of an MRI Scanner

In late 1992 and early 1993 I was fulfilling my internship at Philips GmbH, Forschungsabteilung Technische Systeme Hamburg. Due to my previous experiences with the simulator for the flame-tomography I was assigned the task to write a simulator for an MRI-scanner. The objects in question were very simple head-phantoms like the one to the left, essentially simple, 2D, geometric objects (this was 1992, mind). As its input the program would take a description of the phantom, and as it's output the simulator would generate data which could then directly be fed back into the tomography-software used at Philips; the reconstruction could then be compared with reconstructions of an actual phantom in an actual MRI-scanner.

Simulation and Tomographic Reconstruction of a Flame

In late 1991, early 1992 I was working on this project as my "Studienarbeit" (3rd year project, BS-thesis). The goal was to get a tomographic reconstruction of the temperature distribution inside a burning flame (assumed rotationally symmetric) from 4 images taken with narrow-band spectral filters. The trick here is to model the flame both as emitting but also as absorbing light. At that time no hardware was available for the task (albeit ordered), and it was my task to write software which would simulate the burning flame, and also to write software which would do the tomographic reconstruction.

One thing which we learnt from doing the simulations (not that one couldn't have found out applying some little error-propagation and common sense) was that we would need a camera with either a 12 Bit linear or 8 Bit logarithmic resolution - both something not readily available in 1991, which ultimately doomed this project.
Find out more...


1/99-8/99, 2/00-9/06
Hamburg University's Cognitive Systems Group, Prof. Bernd Neuman
The University of Freiburg's Institute of Pattern Recognition and Image Processing, Prof. Hans Burkhard
The TU Hamburg-Harburg's now somewhat obsolete Computer Vision group (TI-1), Prof. Hans Burkhard
7/93-1/94, 4/94-5/94, 7/95-10/95
The University of Oxford's Robotics Research Group, the Visual Geometry Group, Dr. Andrew Zisserman (my old homepage)
10/92-1/93, 4/93-5/93
Philips GmbH, Forschungsabteilung Technische Systeme Hamburg, Dr. Karsten Ottenberg



Utcke, Sven: Error propagation in geometry-based grouping, Albert-Ludwigs-Universität Freiburg, Institut für Informatik, Fakultät für Angewandte Wissenschaften, Doktorarbeit, 2006
[BibTeX] [Abstract] [Pdf, 11MB]

Porter, Dorrit; Werner, Peter; Utcke, Sven; Fachbereich Informatik, Universität Hamburg (Hrsg.): Ancient Ceramics: Computer aided Classification, Mitteilung. FBI-HH-M-338/06, 2006
[BibTeX] [Abstract] [pdf, 401KB] [ps.gz, 455KB]


Utcke, Sven: Error-Bounds on Curvature Estimation. In: Scale Space: Springer-Verlag, Berlin, 2003, S. 657--666
[BibTeX] [pdf, 200KB] [Abstract, 1.0KB]

Utcke, Sven; Fachbereich Informatik, Universität Hamburg (Hrsg.): Comparison of Different Approaches for the Calculation of Projective Symmetry or the Axis of a SHGC, Bericht. FBI-HH-253/03, 2003
[BibTeX] [Abstract] [pdf, 1.1MB]

Utcke, Sven; Fachbereich Informatik, Universität Hamburg (Hrsg.): Error-Bounds on Curvature Estimation, Bericht. FBI-HH-252/03, 2003
[BibTeX] [Abstract] [pdf, 110KB]

Utcke, Sven; Zisserman, Andrew: Projective Reconstruction of Surfaces of Revolution. In: Michaelis, B. (Hrsg.); Krell, G. (Hrsg.): 25. DAGM-Symposium Mustererkennung. Lecture Notes in Computer Science. Bd. 2781: Springer-Verlag, Berlin, 2003. - ISBN 3-540-40861-4, S. 265--272
[BibTeX] [pdf, 660KB] [Abstract, 1.0KB]


Utcke, Sven: Grouping based on Projective Geometry Constraints and Uncertainty. In: Proceedings of the Sixth International Conference on Computer Vision: Narosa Publishing House, New Delhi, 1998. - ISBN 81-7319-221-9, S. 739--746
[BibTeX] [ps.gz, 1.3MB] [Abstract, 1.0KB]

Luo, An; Tao, Wenjing; Utcke, Sven; Burkhardt, Hans; Albert-Ludwigs-Universität, Freiburg, Institut für Informatik (Hrsg.): MOVIS: Über die Entwicklung eines ersten Prototypen einer Blindenbrille, Interner Bericht. 3/98, 1998
[BibTeX] [ps.gz]


Utcke, Sven; Technische Informatik I, TU-HH (Hrsg.): Grouping based on Projective Geometry Constraints and Uncertainty, Interner Bericht. 1/97, 1997
[BibTeX] [ps.gz, 1.9MB] [Abstract, 1.0KB]


Mundy, J.; Liu, A.; Pillow, N.; Zisserman, A.; Abdallah, S.; Utcke, S.; Nayar, S.; Rothwell, C.: An Experimental Comparison of Appearance and Geometric Model Based Recognition, 1996

Mundy, J.; Liu, A.; Pillow, N.; Zisserman, A.; Abdallah, S.; Utcke, S.; Nayar, S.; Rothwell, C.: An Experimental Comparison of Appearance and Geometric Model Based Recognition. In: Proc. Object Representation in Computer Vision II. LNCS 1144: Springer-Verlag, 1996, S. 247--269
[BibTeX] [Postscript, 1.6MB] [Abstract, 1.5KB]


Pillow, Nic; Utcke, Sven; Zisserman, Andrew: Viewpoint-Invariant Representation of Generalized Cylinders Using the Symmetry Set. In: Image and Vision Computing 13 (1995), Nr. 5, S. 355--365
[BibTeX] [ps.gz, 850KB] [Abstract, 1KB]

Zisserman, Andrew; Mundy, Joe; Forsyth, David; Liu, Jane; Pillow, Nic; Rothwell, Charlie; Utcke, Sven: Class-Based Grouping in Perspective Images. In: Proceedings of the Fifth International Conference on Computer Vision, 1995, S. 183--188
[BibTeX] [ps.gz, 735KB] [Abstract, 1.8KB]

Müller, Volker; Utcke, Sven: Advanced Quality Inspection through Physics-Based Vision. In: Proc. of the the International Symposium Machine Vision in the Industrial Practice, 1995
[BibTeX] [Postscript, 1.2MB] [Abstract, 1.5KB]


Pillow, Nic; Utcke, Sven; Zisserman, Andrew: Viewpoint-Invariant Representation of Generalized Cylinders Using the Symmetry Set. In: British Machine Vision Conference, 1994

Mundy, J. L.; Huang, C.; Liu, J.; Hoffman, W.; Forsyth, D. A.; Rothwell, C. A.; Zisserman, A.; Utcke, S.; Bournez, O.: MORSE: A 3D Object Recognition System based on Geometric Invariants. In: Kaufmann, Morgan (Hrsg.): Image Understanding Workshop, 1994, S. II:1393--1402
[BibTeX] [ps.gz, 216KB] [Abstract, 0.5KB]

Utcke, Sven: Transfer and Invariants of Surfaces of Revolution, Technische Universität Hamburg-Harburg, Diplomarbeit, 1994
[BibTeX] [ps.gz, 1.6MB] [pdf, 1.4MB]