NSF-BSF: HCC: Small: Computational Imaging with Speckle Correlations for Material Analysis

Light scattering materials are ubiquitous: from biological tissues and minerals to the atmosphere and clouds, cosmetics, and many industrial chemicals. When viewed under coherent imaging conditions such as laser illumination, these materials create "noisy" images containing small spurious spots of color known as speckle, which severely degrade image quality.

Despite their seemingly random nature, speckle images have strong statistical properties that are highly informative of the material producing them, and can be used to enable remarkable imaging capabilities (for example, being able to see through or focus inside a material). However, realizing these capabilities in practical application settings where scattering is important (tissue imaging, fluorescence microscopy, remote sensing, material fabrication) remains a challenge due to our incomplete understanding of speckle statistical properties.

This project will use techniques from computer vision and computer graphics to greatly enhance our understanding of these properties, and significantly expand the scope of possible applications, thereby having transformative impact in areas such as medicine and material science. The project will additionally create a new point of convergence between vision, graphics, optics, and imaging, establishing new research directions and methodological approaches within and across these areas.

Achieving the project's goals will require developing a better understanding of the limitations inherent in speckle statistics, and using this knowledge to invent better imaging algorithms. To date, efforts towards this direction have been hindered by difficulties in simulating speckle effects stemming from the complexity of the wave-optics and multiple-scattering phenomena underlying them.

This project will change this state of affairs through three tightly coupled research thrusts: 1) Simulation The project will first develop Monte Carlo rendering algorithms that efficiently simulate physically accurate speckle images and different types of statistics (spatial, temporal, spectral). These tools will be used to perform a thorough qualitative and quantitative investigation of the statistical properties of speckle images, with emphasis on application-relevant settings.

Phenomenological discoveries will be backed by theoretical analysis of the underlying physics. 2) The project will develop new algorithms for speckle-based imaging through scattering, using the insights from the exploration of speckle statistics. These algorithms will be designed to specifically exploit additional statistical structure in speckle images that currently remains untapped.

The improved performance of the developed algorithms will be demonstrated through experiments on tissue phantoms and under conditions emphasizing fluorescence imaging applications. 3)The project will develop computational imaging systems that use the rich information available in speckle images to characterize the optical scattering properties of material samples (for instance, for material quality control). The use of speckle measurements will endow these systems with a combination of accuracy, efficiency, and generality that is not available in existing scattering acquisition technologies.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.