Computer Vision & Pattern Recognition
The purpose of the course is to introduce basic problems and notions in image processing, computer vision, and patter recognition though a common geometric framework and present some classical, industry-standard and state-of-the-art methods through this framework. The course uses tools from differential geometry, calculus of variations, and numerical optimization to address problems such as image recovery (denoising, impainting, deconvolution), filtering (adaptive diffusion, bilateral and non-local means filters), 3D structure reconstruction (shape from shading, stereo, photometric stereo); and rigid and non-rigid similarity and correspondence (iterative closest point methods, multidimensional scaling, Gromov-Hausdorff distance). The emphasis is made on both formulating a rigorous mathematical model of the problem and developing an efficient numerical method for its solution, with hands-on programming exercises that solve real-world problems.
Geometric Deep Learning, Machine Learning
- R. Kimmel, Numerical geometry of images, Springer 2003
- A. M. Bronstein, M. M. Bronstein, R. Kimmel, Numerical geometry of non-rigid shapes, Springer 2007