The objective of this research is to develop a hierarchical representation based on simultaneous sparse coding and explore its potential applications in energy-efficient sensing. The newly constructed hierarchical representation is expected to offer powerful prior models suitable for Bayesian reconstruction of images from noisy or incomplete samples. Two novel applications related to energy-efficient sensing will be studied: low-illumination imaging and super-clarity imaging where the proposed computational approach can help strike a better tradeoff between the cost (energy) and the quality of acquired images.
The intellectual merit of this project includes: 1) hierarchical extension of simultaneous sparse coding will require a new theoretical framework operating with the fewest artificial mathematical structures (e.g., not even the definition of a metric is introduced); 2) the two applications of energy-efficient sensing are new frontiers of computational imaging with potential high impact on both industry (e.g., consumer electronics, mobile devices) and scientific community (e.g., super-resolution microscopy, computational astronomy).
With respect to broader impacts, the PI will make special effort to broadly disseminate the research results at a large scale to promote experimentally-reproducible research in sparse representations and computational imaging. This project will broaden the participation of underrepresented groups by promoting geographic and gender diversities within the NSF program. The proposed research is beneficial to the general public by bringing images of higher quality to people's daily lives at a lower cost.
