List of Past Events
Visual patterns with matching subband statistics and higher order image
Dr. Joshua Gluckman
Monday, April 10, 2006, 02:00pm - 03:00pm
Polytechnic University
Statistical representations of visual patterns are commonly used incomputer vision. One such representation is a distribution measured fromthe output of a bank of filters (Gaussian, Laplacian, Gabor, waveletetc). Both marginal and joint distributions of filter responses havebeen advocated and effectively used for a variety of vision tasks. Webegin by examining the ability of these statistical representations todiscriminate between an arbitrary pair of visual stimuli. Examples ofpatterns are derived that possess the same statistical properties, yetare �visually distinct.� The existence of these patterns suggests theneed for more powerful early visual representations. It has been arguedthat the primary role of early vision is the modeling of statisticalredundancy in natural imagery. One of the most striking statisticalproperties of images is scale invariance. In the second part of thistalk, the scale invariant properties of images are examined and a novelimage representation termed the higher order pyramid will be discussed.The representation is tuned to the scale invariant properties of imagesand constitutes a form of �higher order whitening.�
