The Taub Faculty of Computer Science Events and Talks

The talk addresses the analysis of multi-spectral retinal image sets and hyperspectral satellite images. Pixels in such image sets represent mixtures of pure substances, called endmembers. The linear mixture model for hyperspectral imaging assumes that each pixel is a linear combination of the spectra of a collection of endmembers. To demix each pixel, these endmembers must be identified. A novel parallel algorithm to extract endmembers and demix each pixel is presented.

Typical algorithms for endmember extraction do not use that each pixel is a combination of only few endmembers. Unlike others, our algorithm is based on the sparse representation of each pixel. The proposed algorithm solves a series of independent minimization problems with sparsity constraints and then identifies endmembers according to a ranking of the computed coefficients.