Super Resolution for Optical Imaging

High-resolution images are greatly sought after in many commercial and military optical systems because they can store large amounts of information, but detector costs increase significantly with pixel count.This project uses the super resolution technique of compressed sensing to improve the resolution of an imaging system limited by too few pixels. Compressed sensing encodes compressed information into low-resolution measurements allowing the reconstruction of high-resolution images from many low-resolution images. The super resolution imager designed uses compressed sensing to decrease the number of samples needed to perfectly reconstruct an image. With a few low-resolution images modulated by a subpixel code, a linear system is generated and solved for that predicts a high-resolution image using small amounts of measurement data. The prototype imager uses two encoding methods –binary mask-actuator encoding and a digital micro-mirror device, and two reconstruction methods –an exact L1 minimization reconstruction and an alternative, inexact machine-learning-based reconstruction. The prototype imager built achieves a fourfold increase in spatial resolution in both X and Y directions, providing a total high spatial resolution gain of 16 times.