I'm interested in both of these topics independently, and wondering some good applications of comp geom within the domain of vision. What are some nice applications and potential projects of comp geom applied to vision?
Probably the most useful application of computational geometry within computer vision is in the realm of shape analysis/matching and image registration. There's a lot to google here, but you can start with Kendall's book on shape space, and also look at problems of measuring the distance between shapes.
Computer vision research often involves high-dimensional analysis of feature space (e.g. pairwise distances, range queries, nearest neighbors, etc). Classical computational geometry data structures, such as KD trees appear a lot in the CV literature.
I would start into looking space-partitioning data structures and their usage in clustering, vector quantization, and density estimation. H. Samet is an expert in the field: Foundations of Multidimensional and Metric Data Structures