A multichannel optical correlation processor based on volume holographic as
sociative memory in a photorefractive crystal and wavelet transform is prop
osed for human face recognition. Distortions due to shift, rotation, scale,
and partial hiding are studied to understand invariant performance of the
processor. Our results show that shift-invariance and rotation-invariance a
re key problems for practical applications of the processor to human face r
ecognition. Theoretical analysis and simulation conclude that the focal len
gth of the Fourier transform lens is the main factor to affect shift-invari
ance of the processor. Shift-invariance would be improved if the focal leng
th were enlarged. With regard to rotation-invariance, a novel mechanism to
recognize human faces with any rotation angle is proposed and testified by
experiments. The processor is more practical with the improvement of invari
ance. (C) 2000 Elsevier Science B.V. All rights reserved.