CS 236805: Seminar on 3D reconstruction, 2000

Seminar (Advanced topics in CS) 236805

3D reconstruction from theory to practice

Ron Kimmel
Spring Semester 2000
Sunday 12:30-14:30 (Taub CS Bldg.) room 006

April 4: Ayelet Pnueli, Shape from Photometric Stereo.
April 9: Vladislav Rafael, (svladi@techst02) Shape from Stereo (including Ch. 6 Faugeras book).
May 21: Bogomyakov Alexander (alexb@cs), Piecewise Smooth Surface Reconstruction
May 28: Rachel Genusov (rgenusov@hotmail.com), Registration of surface patches
June 4: Asia ???, Pyramidal methods for Shape from Stereo
June 11: (12:30-14:30) Ginzburg Michael (ginzburg@tx) 3D reconstuction from MRI images.

Next lectures:

June 18: (10:30-12:30) Marina Gruzd, (gruzd@tx) Volume reconstuction from MRI images.
June 18: Valery Weissman, (sval@techst02) Stereo Motion: The Factorization Method.

In image analysis and computer vision we often encounter the problem of 3D shape reconstruction from a flat data like one or two intensity images. The goal of this course is to introduce geometric algorithms and numerical methods, and apply them to 3D shape reconstruction problems in Computer Vision and Graphics. We explore new methods in computer vision for 3D object segmentation, signatures, and 3D shape reconstruction from `flat’ images like

1. Random dot autostereograms, generation and reconstruction.
Those noisy looking images of repeating patterns from which a 3D shapes (sometimes) pops out.
2. Shape from shading.
3. Shape from stereo.
4. Shape from photometric stereo.
5. Shape from structured light.
6. Shape from laser scanning.
6. Shape from texture.
7. 3D shape reconstruction in volumetric medical data.
8. Segmentation in volumetric data.
9. Shape from motion.
10. Some theory and practice of invariants of 3D manifolds.

Course requirements: a frontal lecture & a short technical report & implementation, & active participation.

Preliminary requirements: Calculus II & Linear Algebra & ({intro.2} image processing || computer vision || computer graphics || numerical analysis || intro. to signal proc. || differential geometry).