Xavier Tizon, 2002-11-13 1
Computerised Image Analysis MN2, HT02
Exercise 2: Active Shape Models and Snakes
Active Shape Models (ASMs) and Snakes are two types of computer based methods used to find
object boundaries. Both ASMs and Snakes use a priori information about the searched shape, and
are hence also referred to as knowledge based methods. However, although the end result may be
similar, the way ASMs and Snakes achieve the goal of locating an object’s boundary is quite
different. The purpose of this exercise is to give a hint of the difference between the methods and to
give some practical experiences with ASMs and Snakes.
Formality
• I would prefer you to work in groups of two persons. That way, you can discuss with each other and solve
difficulties more easily. It should be possible to work from any place where Matlab is available.
• You should prepare a written report, one for each group. Remember when you write the report that it should
show that you have learned about the methods you have worked with. I am not that interested in how well
you solved the particular problem, as long as you describe precisely what you have understood.
• Hand in the report to Ingela or to me, or send it by email to xavier@cb.uu.se. You are very welcome to write
the report in English, but I can read Swedish too, or French if you like J.
• As the first part is more passive, I would like you not to spend more than the third of your time on this one,
and concentrate more on the second part, that requires more thinking.
• Deadline: at the latest at the day for the written exam.
Part 1. Active Shape Models (ASM)
The ASM package you will use was developed by Ghassan Hamarneh at the Image Analysis Group,
Department of Signals and Systems at Chalmers University of Technology, to which I made some personal
modifications. The ASM model implements a statistically constrained Point Distribution Model. The user trains
the model on several examples by choosing landmark points on a set of example images. A Principal Component
Analysis is then derived from the training stage and gives the modes of variation of the model. These parameters
are used to help the model fit the final dataset, by allowing model variability only along the principal variation
modes. It is based on an article by Cootes et al ("Active Shape Models – Their Training and Application",
Computer Vision and Image Understanding, Vol.61, No.1, pp.38-59, 1995) that you can copy from me if you
need. The description given in the course book (Sonka, pp. 380-390) should nevertheless be enough to do the
lab.
Getting started
To start the ASM program:
1. Copy the file http://www.cb.uu.se/~ingela/Teaching/ImageAnalysis/Snakes/asm.tar to your working
directory.
2. Unpack the archive (type tar xvf asm.tar). A main directory named asm will be created.
3. Start Matlab
4. Go in the ASM directory and type asm
- 1
- 2
前往页