Architectures for Compressive Imaging with Applications in Sensor Networks, Adaptive Object Reconstruction, and Motion Detection

Abstract

Computational imaging becomes a cutting edge research area by incorporating signal/image processing as an inherent part of an imaging system. Its civil and military applications include surveillance, automobile, and medical health. The newest branch of computational imaging, compressive imaging emerged in several years back. In-stead of making measurement for each individual object pixel, compressive imaging directly making compressed measurements using optical/opto-electronic devices in data acquisition process. These compressed measurements referred to as features are linear combinations of object pixels weighted by transformation bases. Usingvarious types of signal processing techniques, features are processed for the imaging system final tasks such as reconstruction, detection, and recognition. In this dissertation, three compressive imaging implementation architectures, sequential, parallel, and photon-sharing architectures, are analyzed. Two kinds of applications, object reconstruction and motion detections, are studied using projections including PC (Principal Component), Hadamard, DCT (Discrete Cosine Transformation), Gabor, and random projection. Linear and/or nonlinear algorithms are used for static and adaptive measurements. A webcam based multi-sensor network and a DMD based single detector imaging system demonstrate the dissertation work.