Abstract

I will consider some of the factors to be considered when assessing the capability and limitations of a new imaging modality, using holography in the 60's and 70's as an example. Biography (100 word limit): Joseph W. Goodman received an A.B. Degree from Harvard, an M.S degree and a Ph.D. degree, both from Stanford University. He joined the faculty of the Department of Electrical Engineering at Stanford in 1967, chaired the department from 1989 to 1996, and served as Senior Associate Dean of Engineering until 1999. He retired from Stanford in January of 2001. Dr. Goodman is the author of the books Introduction to Fourier Optics (now in its 3 rd edition), Statistical Optics, Speckle Phenomena in Optics. He has received numerous awards from the I.E.E.E., the A.S.E.E., the O.S.A., the S.P.I.E ., including the highest awards given by the latter two societies.Blind Image Quality Assessment (Blind IQA) is usually synonymous with "no-reference" IQA, viz., without a pristine image available for quality comparison. However, other sources of information are used to design IQA models: human opinion scores, mathematical models of distortion, perceptual models; statistical models of distorted images, and statistical models of images that are not distorted. Given this variety of information sources, many degrees of "blind" IQA can be contemplated resulting in different types of models and algorithms. At one end of the spectrum are IQA models that use a lot of information, employing sophisticated machine learning methods driven by perceptual and natural scene models, subjective data, and distortion exemplars to achieve high performance similar to the best "full reference" (FR) non-blind algorithms. At the other end are models that use very little information other than examplars or models of undistorted images, yet achieve performance comparable to traditional FR metrics such as PSNR. I will discuss the many variables available for designing "blind" IQA models and how we are using them to create IQA algorithms that can be used to monitor image quality in real time, to evaluate and benchmark image processing algorithms, and to optimize image processing, repair, and compression methods.

© 2012 OSA

PDF Article