Joint Design of Optics and Image Processing for Application-Specific Sensors: Overthrowing Old Optical Design Principles in the New Era of Electro-Optics
|Event Date:||March 24, 2011|
|Speaker:||Dr David G. Stork|
|Speaker Affiliation:||Chief Scientist, Ricoh Innovations|
|Contact Name:||Professor Mary Comer
|Open To:||ACCEPTABLE FOR ECE694A
Traditional principles of optical imaging guide the search for designs that yield high-quality optical images (e.g., small point-spread functions and low aberrations). In the new world of computational imaging, optical images are digitally sensed and processed before presented on a display screen and as such only the quality of the final digitally displayed image is important, not that of the intermediate optical image. Simultaneous joint design of the optics and image processing for the end-to-end merit function differs in fundamental ways from traditional optics-only and from sequential optics then image processing methods. The resulting trend is for imaging design to rely less upon the physics of light, optical elements and photons and more upon the information theory of digital signals, image processing algorithms and bits.
In the new world of computational imaging, large PSFs (blurry optical images) that lack zeros in the MTF are superior to small PSFs (sharp optical images) because the information can be recovered through digital Wiener filtering. Moreover, because some optical aberrations are easier to correct through digital processing than others, traditional optical merit functions for comparing aberrations are of little use when designing electro-optical imaging systems. In fact, certain optical aberrations can be advantageous in a digital-optical system: severe chromatic aberration can, through digital processing, yield enhanced depth of field. Because more of the imaging burden is carried by image processing, for a given imaging performance (measured in MSE), these new systems can be smaller, have fewer optical elements, have higher manufacturing yield—and thus lower cost—than imaging systems designed through traditional sequential methods.
These new principles apply not merely to systems employing esoteric optical elements (such as cubic phase plates for enhanced depth of field) nor new imaging architectures (such as lightwave sensing) nor radical designs (such as ultra-thin multi-lenslet integrated cameras) but also to traditional spherical-lens imagers such as a simple singlet and Cooke triplet.
Given the relentless trends...
...away from chemical image sensing toward digital sensing and display
...toward more powerful, less-expensive digital image processing hardware
...toward more application-specific imagers
these new electro-optical imaging design principles are likely to become ever more important.
Dr. David G. Stork is Chief Scientist of Ricoh Innovations and Fellow of the International Association for Pattern Recognition. His eight books/proceedings volumes include Seeing the Light: Optics in Nature, Photography, Color, Vision and Holography, Pattern Classification (2nd ed.), HAL's Legacy: 2001's computer as dream and reality, and three volumes on computer vision applied to the study of fine art paintings and drawings.