Skin Deformation Modelling using image analysis

I am trying to model skin deformation. Currently I have set up where I paint a matrix of dots onto the skin. I have a program that tracks the points and hence I can calculate deformation as a function of time. However, the use of ink on the skin is not an ideal solution.

Firstly, the ink is constantly diffusing into the skin (especially during deformation) so the areas of the points is not consistant. Also, too fine a point matrix (to increase accuracy) will change the mechanical properties of the skin.

Is there any way I could track the natural creases/grooves of the skin? I understand this would be much more complex. Currently all testing I have done is biaxial (no Z-direction deformation) but ideally I would like to measure 3D deformation. I have attached a sample of the type of skin image I am using.

Answer 1

[Image Analyst]

I think for an image of that kind and magnification, you're best off using optical flow. It tracks movement not by looking at isolated markers (like the motion capture industry that is used heavily in the entertainment industry), but by looking at all of the image pixels themselves. For example see some of the papers by this professor: John Barron where he uses optical flow to track storm, movement of plants, etc. Or you Google the subject of optical flow.

[Yasha]

so would optical flow enable me to determine the x,y (and possibly z?) co-ordinates as a function of time?

[Image Analyst]

Yes, if you have images that are a function of time, like from a video or time lapse photography. Basically it gives you little vectors for each pixel telling you where it moved to in the next image.

[Yasha]

Im in my last year at university, and the project I am currently involved in will only last till March. Would you think I would have enough time to develope such a program? I guess there would be similar programs I could base my program on.

[Yasha]

Also, the examples I have seen are where there is a constant background with a moving object. In my case, the whole image will be displaced. Would it still be possible to use optical flow to map the displacement?

[Image Analyst]

Yes it's possible, in fact that's the most common use I would think. But optical flow is not trivial and I don't know your skills so I don't know how much you will be able to accomplish by March.

[Yasha]

Realistically, I would like to be able to analyse a 2D model. I do not have access to the image processing toolbox, is this a major setback?