APPLIED MATH SEMINAR

Speaker: Matthew Hirn, Yale University

Title: Sparse endmember extraction and demixing

When/Where: Tuesday, October 6th, 4:15 PM, AKW 200

Abstract: The analysis of multi- and hyperspectral images is not only important
in geospatial imaging, but also in biomedical applications. Each pixel
represents the mixture of pure substances and one of the major challenges is to
determine these substances, called endmembers, and then demix each pixel.

Standard endmember extraction uses a linear mixing model and varies from
geometric methods to heuristic approaches. N-FINDR determines endmembers as the
vertices of the simplex spanned by the pixel vectors. Pixel purity index is
probably the most commonly used algorithm due to its integration in ENVI.

In this talk we will replace geometric methods with an algebraic approach based
on mixed l2 and l1 norms. We will describe the current state of this approach
both in terms of theoretical underpinnings as well as empirical results.

This talk is based on joint work with Martin Ehler (University of Maryland and
National Institutes of Health).