SVD and clustering with 3D data
I am working on a research project where we have 3D scans of people's backs, and we would like to cluster the back shapes using clinically-relevant dimensions. For example, one set of people might benefit from a certain backrest due to their height.
The backs are currently formatted in uniform grids that when 3D-plotted in Matlab look like backs. My adviser thinks it is possible to cluster the back scans somehow using singular value decomposition (SVD). However, I am having trouble understanding the necessary steps in this process. Do I cluster using a particular matrix in the decomposition (e.g. U, sigma, V) or something else?
I am open to suggestions, so if you know of any methods that might lead us to our goal, I would like to hear them. I have not studied SVD before my adviser suggested I use it for this project, so I would appreciate responses in as elementary terms as you can make them.