Brendan McCane
Current Research
Currently I am involved in several research projects. My main recent focus has been on 2D shape analysis, especially in the context of statistical shape analysis in medical applications and analysing lumbar radiographs and videofluoroscopy images. I also have an interest in large margin classifiers - AdaBoost and support vector machines, and in feature extraction. Also, with a summer student and a colleague in genetics, I've been looking at regulatory gene networks.
I am always on the lookout for applied problems in computer vision and pattern recognition, so if you have any interesting applied problems and you would like to collaborate, please get in touch.
Information for Potential Interns
I get a lot of requests from students wanting to do research internships with our lab. Usually we are able to accommodate interns, but we don't have any money to pay you. So unless you have other sources of funding, or can afford to come anyway, don't bother asking. We can provide equipment, supervision and a great atmosphere though. If you're still interested then by all means send me an email, but please mention that you've read and understood this webpage, and that you realise that we can't pay you. And I really mean we can't pay you - not for accommodation or sustenance or anything.
Potential Student Projects
These projects could be of interest to MSc, PhD students or interns. Note that the University of Otago now has an excellent scholarship program (although quite competitive), and that overseas PhD students need only pay domestic fees. Further, the cost of living in Dunedin is very low. This makes Otago very attractive for PhD students.Regulatory Gene Networks
We have reimplemented and improved upon the Ingenue system for studying regulatory gene networks using partial differential equations. There are many interesting questions left to answer in this work including:
- Given an approximate network, can we deduce one or several networks that produce the desired output?
- Gene networks studied so far appear to be extremely robust to changes in their parameters. That is, they often still work despite widely varying parameters. Given this is true, how can evolution happen since small tweaks in the parameters typically does not change the function of the network?
Learning Image Search Algorithms
Many computer vision problems require some form of search to locate an object in an image. A fair amount of research effort is required to develop algorithms to solve each new problem. This project is aimed at the problem of attempting to learn an appropriate search strategy to solve the problem.
Cephalometric Analysis
Cephalometric analysis is the study of shapes in the human jaw and skull and is often used by orthodontists for diagnosing problems, planning corrective action, and evaluating the results. This project involves developing better techniques for analysing the shapes evident from x-rays of the human skull.
Cascaded Support Vector Machines
Adaboost classifiers can be cascaded to make them run really fast. The question to answer here is whether support vector machines can use a similar cascaded architecture.
Red Blood Cell Classification
I have in my posession a database of electron scanning microscope images of red blood cells. Red blood cells come in several different shape types. A previous student has developed an algorithm that successfully segments the blood into individual cells. The remaining tasks involve developing algorithms to classify each cell according to its shape.