Computer Vision - SS 08

This lecture is concerned with the problem how computers can process and understand images, with a strong focus on recent methods for object recognition and categorization. The lecture will be given in English by Prof. Bernt Schiele. The exercises will be supervised by Micha Andriluka.

Announcements

• Please send me an email (andriluka AT mis DOT informatik DOT tu-darmstadt DOT de) with your RBG login if you want to use Matlab in the computer pool.
• Important! Everyone who takes part in the course should subscribe to the mailing list: https://mail.rbg.informatik.tu-darmstadt.de/mailman21/listinfo/cv
• Please use "Computer Vision" forum provided by the Fachschaft to discuss issues concerning exercises. Forum can be found at www.fachschaft.informatik.tu-darmstadt.de/forum/viewforum.php?f=290 (note that general announcements will still be disseminated primarily through the mailing list).

Lecture Description

The focus of the lecture lies on recent methods for object recognition. The main task is here, given an input image (e.g. a photograph or a frame from a video sequence), to decide what is visible in the image. For this, we can distinguish between two different problems: the identification of known objects in a different context ("Is that my cup / dog / car / etc.?") and the recognition of whole object categories ("Is this a cup / dog / car / etc.?"). The latter problem is particularly challenging, as we cannot assume to have seen every possible object of that category before. The task is therefore to learn what are the characteristics of an object category by which it can be recognized. An important point is that this learning step should not require the user to enter a fixed set of rules into the system, but the computer should learn the category properties by itself, just based on a relatively small number of training images. Topics to be covered in the lecture include: Basic image processing, image filtering, edge and 2D structure extraction Statistical methods for object recognition Interest point operators and local descriptors Part-based methods for object categorization and segmentation Format: V2 + U2 When and Where Lecture: Wednesday, 8:00-9:40, location: S02|02 C110 Exercises: Friday, 11:40-13:20, location: S02|02 C110

Literature

In the last decades, Computer Vision has evolved into a sprawling field with research going into so many directions that no single book can cover them all. A good general introduction to Computer Vision can be found in the book:

D. Forsyth, J. Ponce Computer Vision -- A Modern Approach, Prentice Hall, 2002

However, most of the material presented in this class is the result of very recent research, so it hasn't found its ways into text books yet. Wherever research papers are necessary for a deeper understanding of the lecture topics, we will make them available on this web page.

Lecture and Exercise Plan

Exercises

The exercises will be supervised by Micha Andriluka.

The exercises will consist of short programming projects using Matlab. An introduction to Matlab will be given in the first exercise session.

• Matlab Tutorial
• Getting Started with Matlab
• The Matlab Primer
• Matlab Online Reference Documentation
• A useful Matlab Quick-reference card (in German).

Office hour

Tuesday, 13:00 - 14:00 in S2/02 E102 (exceptions: on 3.06 and 1.07 office hour will be in the room E202; there will be no office hour on 24.06)

Contact

For all questions concerning the lecture, please contact

Last changed: Micha Andriluka, 09.04.2008