Discriminant An ct n from Vie
D.L. Swets · 1996
The method we have been usang is based on our SelfOrganazang Haerarchacal Optamal Subspace Learnang and Inference Framework (SHOSLIF). It uses the theoraes of lanear dascramanant projectaon for automatac optzmal feature selection an each of the internal nodes of 0 SpaceTessellataon Tree. In thas paper, we present our recent study on the applacabzlaty of the approach to varaabalaty zn posataon, sue, and 3D oraentataon. In the work presented here, we requare LLwell-framed” amages as anput for recognataon. By well-framed amages we mean that only a relatavely small vamataon an the sue, positaon, and oraentataon of the objects an the anput images is allowed. We report the experamental results that show the performance daflerence between the subspaces of lanear dascramanant analysas and the prancaple component analysas and the ejfect of usang a tree as opposed to a fiat eagenspace.