This Small Business Innovation Research (SBIR) Phase II research project proposes to develop an adaptive CMOS image sensor that estimates and largely eliminates illumination variations in sensed optical images thus reporting electronic images that are indicative of the reflectance of the viewed scene. By eliminating illumination-induced variations from the raw optical images the proposed sensor will eradicate the vision system's vulnerability to illumination variations and signal loss due to high dynamic range. The core innovation is in a signal processing technique for estimating the illumination field from sensed images. The technique efficiently implements as a dense on-chip massively parallel analog processor distributed among the photo-detectors to produce a reflectance sensitive image sensor. By compensating for illumination, the proposed image sensor inherently addresses the wide dynamic range problem, that routinely causes conventional cameras to over or under expose producing inadequate images. Even when illumination conditions do not saturate an image sensor, the vision system has to account for object appearance variations caused by illumination.
The proposed research has the potential to broadly impact computer vision performance and reliability. Most present and future vision applications including automotive, biometric, security, and mobile computing applications operate in unconstrained environments and have to cope with unknown and widely varying illumination conditions. Image sensors are rapidly finding their way into people's cars, cell-phones, personal digital assistants, medical and diagnostic equipment, automated drug discovery, cutting edge security, surveillance and biometric systems.
