Compressive sensing is a new technology that allows one to acquire data with high information content, based on measurements that are much fewer in number than would be expected by Nyquist theory. This is achieved by leveraging signal models that go beyond signal bandwidth, to exploit the fact that signals of interest typically reside in a low-dimensional subspace. This includes exploitation of concepts like sparsity, unions of subspaces, and low-rank signal models, as well as nonlinear signal-recovery methods. In this talk we examine how modern statistical tools may be used to achieve remarkable levels of compression directly at the measurement point, with demonstrations based on new hyperspectral and video cameras. The talk will examine the fundamental mathematics and statistics, and show results based on real cameras we have developed and tested.
Researchers should cite this work as follows:
Birck Nanotechnology Building, Room 1001, Purdue University, West Lafayette, IN