A new method to recognize partially visible two-dimensional objects by
means of multiscale features and Hopfield neural network was proposed
. The Hopfield network was employed to perform global feature matching
. Since the network only guarantees to converge to a local optimal sta
te, the matching results heavily depend on the initial network state d
etermined by the extracted features. To acquire more satisfactory init
ial matching results, a new feature vector was developed which consist
s of the multiscale evolution of the extremal position and magnitude o
f the wavelet transformed contour orientation. These features can even
be used to discriminate dominant points, hence good initial states ca
n be obtained. The good initiation enables our proposed method to reco
gnize objects even heavily occluded, that cannot be achieved by using
the Nasrabadi-Li's method. In addition, to make the matching results m
ore insensitive to the threshold value selection of the network, we re
place the step-like thresholding function by a ramp-like one. Experime
ntal results have shown that our method is effective even for noisy oc
cluded objects. Copyright (C) 1996 Pattern Recognition Society.