Contract W911NF-12-1-0529 (University of California - Berkeley).

摘要

In this STIR project, we have achieved three major contributions in the application of sparse and low-rank representation in geometric 3D modeling of urban structures and high-dimensional pattern recognition at large. First, we proposed a novel algorithm to effectively detect geometry-rich low-rank image patterns in natural images. Second, we extended a sparse representation- based classification framework to the small-sample-set scenario. Finally, our research on accelerating the speed of sparse optimization solvers was accepted for publication.