Constraint optimization neural network for adaptive early vision

B. Furman: 7#,62 Boelt~r Hall, Dept. of Elec~cal Engineering, U.C.L.A., Los Angeles, CA 90024 J. Limg: Eastman Kodak, 1669 Lake Ave.., Roc.hest~, MN 14650 H. $zu: Naval Ru~uv.h Labs, Code 5756, WlhiniPmt DC 20375 N e ~ o~w_ o ~ ~ ra.t~ .~__p~em.~s for ~ v mions .of ca2~..nt ~ ~ Wy d~:l"ib¢ here a ~ ~ N~'al.Network for .~Miv¢ Early VlSlOa ~.. V.). ~ . The d L ~ l ~ nullfiplim of the onnmmnt Hlmfllt0niaa m v a r i a ~ dm can be . .a~Zed .to opml~m, glo~i. ~ ~ t ca tFe " extraction. Examples of global parameters ~e umc average overall intensity (useful fc/adalZiVe dacslmlding), numt~r of edges, etc. This work is based on models by Poggio, Koch, Mam~uin, and others who ~ ~ t i o n . ~ to E. V. proSms, offir m d ~ ~.L_-~_ ~ ~ bo~ a m ~ t m a ~ comzi= _~m21a~e ~¢~.U o C m m m m ~ N ~ . ~ m ~ n . m t i ~ ~ m / b o t t m up .a.ot~on, and gond ¢onvegeace p, wertm. T~ syam ts saucund f . d~-~t