Research

Thrust R2

Physics-Based Signal Processing & Image Understanding

Physics-Based Signal Processing and Image Understanding creates and verifies inverse algorithms to infer subsurface details from measurement of above-surface sensors. This thrust focuses on optimizing the entire detection system, from the sensor data to the information desired, particularly addressing problems where the map from data to decision is non-linear or multi-modal. The goal is to identify common mathematical structures and develop general approaches that are applicable across diverse application domains. The work is described under four areas, representing distinct problem classes and information extraction strategies. Each of the areas is developing algorithms that are broadly applicable across different applications in these problem classes. The four areas are summarized below:

• Multi-View Tomography (MVT) is concerned with problems where individual sensors capture integrated properties of overlapping areas of the observed subsurface region. These problems arise in many Center applications, ranging from ground penetrating radar subsurface imaging to Electric Impedance Tomography.
• Localized Probing and Mosaicing (LPM) is concerned with problems where individual sensor information reflects properties of a highly localized sub-region of the subsurface problem of interest. In these problems determination of the global properties of the material requires registration and fusion of the multiple sources of localized information. Current applications which require LPM strategies involve retinal subsurface imaging, underwater imaging with sidescan sonar and strobe video, and confocal microscopy.
• Multi-Spectral Discrimination (MSD) is concerned with problems where sensors collect information on the observed problem of interest across multiple cross-registered spectral bands. Properties of the subsurface volume of interest must be inferred from fusion of the spectral information. Applications that require MSD strategies are skin and brain imaging, underwater quantitative imaging from airborne or satellite based hyperspectral sensors, and multispectral optical biopsy for cancer identification.
• Image Understanding and Sensor Fusion (IUSF) aims to extract useful information from the images generated by subsurface inverse problems, such as the underlying object structure contained in a subsurface environment. In many cases, combination of diverse sources of information, obtained at different times and with different modalities, is needed to characterize the structure of the subsurface phenomena under observation. Applications that require IUSF include tumor detection and localization in low-contrast imagery, shape estimation for buried object classification, multi-sensor fusion for coral reef monitoring, multi-modal sensor fusion in breast imaging, and multi-mode microscopy.

Contact Thrust 2 Leader David Castanon at dac@bu.edu for more information.