High quality night vision and infrared cameras are critical for a variety of tasks ranging from security in banks and museums to observing nocturnal animals to military troop safety. In the past, improving the quality of photographs meant using ever larger camera lenses. However, in these and other applications it is important for cameras to be as small as possible.
Did you know? The digital camera market is growing at 50% per year, and there are already over 30 million digital cameras in homes and offices worldwide. Source: InfoTrends
Duke has been working on this challenge since 2003 under the Compressive Optical MONTAGE Photography Initiative (COMP-I), a program sponsored under the DARPA MONTAGE program. The project is led by electrical and computer engineering (ECE) Professor David Brady and includes ECE Duke Faculty Nikos Pitsianis and Rebecca Willett as well as Computer Science Professor Xiaobai Sun. The project team also includes researchers from Raytheon Company, Digital Optics Corporation, Michigan Tech, UNC Charlotte and the University of Delaware.
The main focus of COMP-I is a next generation of ultra-small infrared cameras. At DARPATech 2007 in August, the team is demonstrating a palm sized, ultra thin camera that uses a grid of nine 6.8 mm thick lenses to capture nine different low resolution photographs. The lenses are 4 times thinner and 50 times smaller in mass than a conventional night vision camera.
Then they use software to carefully merge that information into a single digital super-resolution image of the scene. The software uses advanced algorithms to correct for sensor distortion and data losses that cause grainy, blurred images. This research is a significant demonstration of the potential for advanced processing algorithms in compressed imaging applications. Continuing development will focus on reductions in overall microimager volume and battery power.
The basic idea of compressive imaging is that one can used advanced sampling and image interpolation algorithms to produce more image pixels than one measures. This concept is particularly compelling for spectral imaging systems. A spectral image may involve 10-100 spectral channels per spatial pixel. Spectral imaging enables optical systems to identify molecular components in images for biomedical and security applications. The COMP-I team has been particularly successful over the past year in demonstrating compressive imaging in unique single frame (i.e. video rate) cameras.
Duke Imaging and Spectroscopy Program: disp.duke.edu
David Brady web site: ece.duke.edu/faculty/david-brady
Xiaobai Sun web site: cs.duke.edu/people/faculty/?csid=37