Traditional approaches to visual information processing have relayed on CCD/CMOS cameras, analog-to-digital conversion and power-hungry digital signal processors. These approaches are very complex, consume lot of power and often prevent real time extraction of relevant visual information. In order to alleviate some of these problems for high image data rate processing, I will discuss several computational image sensors capable of extracting visual information at the sensor level. I will approach the subject of integrated computational sensors from the information flow perspective. First, I will introduce a novel current mode pixel, which operates with reduced number of transistors per pixel. The new imaging pixel allows for high resolution and low noise current mode imaging. The fundamental noise limitation of this novel pixel will be presented and theoretical noise models will be discussed.
The second imaging sensor will present low power analog techniques for focal plane spatiotemporal image processing. This imaging sensor borrows essential ideas from biology and implements most of the computation in the analog domain. Most of the processing on the raw image is performed during read out, avoiding the power consumption problems and complexity of digital circuits and ADCs. Applications for this focal plane image sensor include motion detection and wave front phase correction.
The final imaging sensor will present a novel sensory system capable of extracting polarization information in real time. Polarization is one of the fundamental properties of light which has been ignored in traditional image sensors. Our research efforts uniquely combine advancements in polymer technology, nano-fabrication and CMOS imaging technology in order to create the first high resolution polarization image sensor. Applications for the polarization image sensor will be presented.
I will conclude with a discussion on image processing architecture implemented in a novel 3-D Silicon on Insulator (SOI) stack fabrication process.