Towards Flexible Sheet Cameras

Northwestern University will act as a subcontractor for the ONR project entitled “Towards Flexible Sheet Cameras”. The Northwestern team will be lead by Professor Oliver Cossairt. The team will be responsible for the following tasks relating to the analysis and development of sheet camera sensors. We will perform a detailed analysis of the optical characteristics (locus of viewpoints, field-of-view, effective focal length, depth-of-field) of the lens-less flexible cameras. We will analyze of trade-off between system blur and signal-to-noise-ratio. Since no optical elements are involved, this will be a purely geometric analysis. We will perform a detailed analysis of the optical characteristics (locus of viewpoints, field-of-view, effective focal length, depth-of-field) of the lens-based flexible cameras. We will analyze of trade-off between system blur and signal-to-noise-ratio. This will involve theoretical analysis of the effects of deformation in optics on the captured images, as well as detailed simulations using Zemax optical design software. Custom software may also be written to assist in simulation of image quality. The goal of this study will be to explore the space of deformation of optics for the purpose of engineering the system sampling rate in a manner which optimally captures the scene information (minimize aliasing and blur). Deblurring and anti-aliasing algorithms will be designed. Deblurring and anti-aliasing will be performed on simulated images, and the quality of deblurred and anti-aliased images will be characterized in terms of the Signal to Noise Ratio (SNR) of the recovered image. The computational efficiency of the algorithms will be studied in anticipation of real-time deblurring/anti-aliasing of 5-10 megapixel images. Tests will be performed for software implementations on various platforms, including GPGPU implementations, which potentially offer a significant Cost/Performance advantage over conventional CPU implementations. The images captured by a sheet camera will almost always be non-perspective images, which are difficult to interpret for humans. We will develop algorithms that map images captured by the sheet camera to perspective images and other formats that are lends themselves to human perception. We plan to use several types of representations including multi-perspective, equi-resolution and collage images. Our goal is to provide the user a range of mapping options to choose from.