18.S096 | Fall 2015 | Undergraduate

Topics in Mathematics of Data Science

18-s096f15.jpg

Description:

Given noisy rotated copies of an image (corresponding to vertices of the depicted graph), the goal is to estimate the original image (or answer: who is in this picture?). Before averaging the images to denoise them, one has to estimate the unknown rotations they were subject to. By comparing some pairs of images (corresponding to edges of the depicted graph), it is possible to estimate the relative rotations between them. The problem of recovering the rotation of each image from these relative rotation estimates is an instance of Angular Synchronization. Image by Dr. Afonso Bandeira.

Alt text:
Twenty rotated copies of a photo with noise added.
Caption:
Given noisy rotated copies of an image (corresponding to vertices of the depicted graph), the goal is to estimate the original image (or answer: Who is in this picture?). Before averaging the images to denoise them, one has to estimate the unknown rotations they were subject to. By comparing some pairs of images (corresponding to edges of the depicted graph), it is possible to estimate the relative rotations between them. The problem of recovering the rotation of each image from these relative rotation estimates is an instance of Angular Synchronization. (Image by Dr. Afonso Bandeira.)
Credit:
Image by Dr. Afonso Bandeira.
Twenty rotated copies of a photo with noise added.

Course Info

Instructor
Departments
As Taught In
Fall 2015
Learning Resource Types
Problem Sets
Lecture Notes
Instructor Insights