Color-based object recognition

The purpose is to arrive at recognition of multicolored objects invariant t o a substantial change in viewpoint, object geometry and illumination. Assu ming dichromatic reflectance and white illumination, it is shown that norma lized color rgb, saturation S and hue II, and the newly proposed color mode ls c(1)c(2)c(3) and l(1)l(2)l(3) are all invariant to a change in viewing d irection, object geometry and illumination. Further, it is shown that hue H and l(1)l(2)l(3) are also invariant to highlights, Finally? a change in sp ectral power distribution of the illumination is considered to propose a ne w color constant color model m(1)m(2)m(3). TO evaluate the recognition accu racy differentiated for the various color models, experiments have been car ried out on a database consisting of 500 images taken from 3-D multicolored man-made objects. The experimental results show that highest object recogn ition accuracy is achieved by l(1)l(2)l(3) and hue H followed by c(1)c(2)c( 3) normalized color rgb and m(1)m(2)m(3) under the constraint of white illu mination. Also, it is demonstrated that recognition accuracy degrades subst antially for all color features other than m(1)m(2)m(3) With a change in il lumination color. The recognition scheme and images are available within th e PicToSeek and Pic2Seek systems on-line at: http://www.wins.uva.nl/researc h/isis/zomax/. (C) 1999 Pattern Recognition Society. Published by Elsevier Science Ltd. All rights reserved.