Home page of Michael van Ginkel

Contact information

This page lives at the Quantitative Imaging Group (formerly Pattern Recognition Group) at Delft University. This is were I got both my Ingenieur degree (=Masters) and PhD. It is convenient for me to hold on to this page -- so that is what I do. This is to our mutual benefit, especially since I am still actively involved in the DIP.

Dr. Ir. Michael van Ginkel
Measurement Science
Unilever Research Colworth,
Colworth House Sharnbrook
MK44 1LQ Bedford
United Kingdom 		Recipe for my e-mail address
1. It is "Michael", followed by a dash ("-")
2. then "van", followed by a dot (".")
3. then "Ginkel"
4. then "@", followed by "unilever dot com"
5. If you can't figure this out, you shouldn't be using e-mail

Research interests

Image Analysis (the black art of extracting information from images). In particular I am interested in:
  • Measurement (orientation, curvature) of structures and objects in images
  • Characterisation (anisotropy, order, organisation) of structures and objects in images
  • Filtering, feature detectors, etc...
  • We pay special attention to discretisation issues. It is natural to design image analysis operators in the continuous domain. But what are the consequences of carrying these over to the discrete domain?
The tools we use to achieve this are mostly based on signal processing and differential geometry.

Coming up with a technique is one thing, whether it works on real-world images is another. We have focussed our work on images from sources relevant to various scientific disciplines, i.e. physics, materials science, food science, geology, etc... using modalities such as (confocal) microscopy, MRI, CT, TEM and various others. Photographs and video streams are out.

My PhD thesis (2002) is mostly concerned with a construction called orientation space which is especially suited to structures which are characterised by more than one orientation (e.g. intersections). The special "thing" about orientation space is how to set it up properly. The proper way is through so-called "steerable filters". In combination with a Radon transform, to which orientation space is, in fact, closely related, we have created a robust curve detection scheme. The latter is quite computationally demanding though. Since then, we have extended our work on a sampling criterion for the Radon transform (in its role as a curve detector). Curvature estimation is another thread in our work: the unique feature of our estimator is that it works on curved patterns rather than individual curves or object outlines. Characteristic for our approach is that it makes use of the fact that, in case of a pattern, an entire neighbourhood of information is available, allowing for a much more robust estimate. In practice, the estimator works well on lines and object outlines as well.

Have a look at my publications if you want to find out more [you know you want to!]

Boring page eh? I still hope to spice it up one day and provide some illustrated examples of my work.

Research links

My publications
I recommend the following books
How did my research fit in the group?
Image Analysis and Pattern Recognition links

Teaching

Not really relevant anymore, but have a look anyway.

Miscellaneous

I dabble in software for Image Analysis. See our DIPimage/DIPlib site. You may also want to have a look at bits 'n pieces I have collected about programming.

IGNORE THIS: The following is a fictitious address to annoy e-mail address harvesters karibou@sahara.desert.zz