Collaborative Research: Accurate, Efficient and Robust Computational Algorithms for Detecting Changes in a Scene Given Indirect Data

Dartmouth College, Hanover NH

Investigators

Abstract

Detecting change from a temporal sequence of collected data is important in a wide variety of applications, including speech recognition, medical monitoring, credit card fraud detection, automated target recognition, and video surveillance. In applications such as medical monitoring, it is very important to find where the change occurs. In other applications, such as video surveillance, the type of change, e.g. the movement or insertion/deletion of an object of interest, is also critical. While detecting such changes from direct data (e.g. images already formed) has been well studied, there are many applications, such as magnetic resonance imaging (MRI), ultrasound, and synthetic aperture radar (SAR) where the temporal sequence of data are acquired indirectly. The typical approach to detecting changes in these applications would be to first form the image or signal of interest. As a consequence, information that is stored in the indirect data that may be valuable to detecting change is often lost. Therefore, this project seeks to develop accurate, efficient, and robust computational algorithms for detecting changes in a signal or image from a given temporal sequence of indirect data without first reconstructing the signal or image of interest. Additionally, the project seeks to incorporate the change information to develop better image and signal reconstruction algorithms. Both graduate and undergraduate students will be involved in the research investigations to enhance their career preparation in science and engineering. The participants will apply these new techniques on publicly available data sets, notably obtained for MRI, ultrasound, and SAR applications. The PIs will employ tools in frame theory, optimization, and statistics to develop and rigorously analyze new change detection and image/signal recovery algorithms. Specifically, the PIs will address the following technical issues in the proposed work: (1) the incorporation of prior information with appropriate mathematical/statistical formulation in the model; (2) the extraction of rotation/translation of an object from a sequence of indirect data; (3) model parameters tuning through statistical analysis; (4) the employment of intra- and inter-signal correlations in the recovery algorithms; (5) the design of distributed algorithms for the resulting large-size optimization model. 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.