Informatisches Kolloquium SoSe 2002
Informatisches Colloquium Hamburg
|
Wenn nicht anders angegeben, finden die Vorträge montags um 17.15 Uhr im Informatikum, Konrad-Zuse-Hörsaal, Gebäude B, Vogt-Kölln-Str. 30, Hamburg-Stellingen statt. |
15.4.2002 |
Prof. Mads Nielsen Feature-Based Image Analysis
According to Marr's paradigm of computational vision the first process is an extraction of relevant features. A stream of artificial vision research has followed this approach. The aim of this work is to quantify and characterize the information in the image that features carry and in this way indirectly validate this concept of feature-based image analysis. The main conclusion is that a reasonably low number of features characterize the image to such a high degree, that nearly all information in the image is revealed. Many images in the space of all images may exhibit the same features. All images exhibiting identical features form a metameric class.The strategy in this work is to define metameric classes of images and to examine the information contents in a canonical least informative representative of this class. Algorithms for identifying these least informative representatives are given. Finally, optimal feature detectors localizing the most informative points (feature-points) in images are derived on the basis of definitions of statistics of images. This leads to new derivations of well known feature detectors and new scale-normalisations of these. |
Kontakt
|
Prof. Dr. Hans-Siegfried Stiehl Telefon +49 40 42883 2453 (stiehl"AT"informatik.uni-hamburg.de) |