Spatiotemporal Imaging Exploiting Structured Sparsity

Overcoming the conventional limits of spatiotemporal imaging by applying compressed sensing and sparse representations to reduce the amount of data acquired while maintaining high image resolution.

Air Force Research Laboratory, APO AP 96338-5002

Spatiotemporal imaging contains a large class of imaging problems, which involve collecting a sequence of data sets to resolve both the spatial and temporal (or spectral) distributions of some physics quantity. This capability is exploited in numerous different fields such as remote sensing, security surveillance systems, astronomical imaging, and biomedical imaging. One typical example is hyperspectral imaging, which is a powerful technology for remotely inferring the material properties of the objects in a scene of interest. Ultrasonic and thermal imaging are other important examples of spatiotemporal imaging where high spatial resolution is needed for urban planning, military planning, intelligence and disaster monitoring and evaluation.

While spatiotemporal imaging has great potential, acquiring and processing data comes with significant practical challenges. First, spatiotemporal images are extremely high-dimensional which limits fast data acquisition. Second, physical design constraints of the acquisition devices (such as size and weight limitations of the satellite) limit attaining higher spatial resolution. As a result, there is a tradeoff between spatial and temporal (or spectral) resolution when designing a system. For example, expensive multiple detector arrays are usually required for recording multispectral bands. By lowering the spatial resolution of each detector, more bands may be contained on the sensor for the same cost. Similar trade-offs are observed in many other spatiotemporal imaging modalities.