scene. Depth information is lost and objects frequently overlap.
Vision requires a large amount of memory and many computations. For
an image of 1000 X 1000 picture elements, even the simplest operation
may require 108 operations. The human retina, with 108 cells
operating at roughly 100 hertz, performs at least 10 billion
operations a second. Thus, to recognize objects at a rate even
closely resembling human vision, very special processor technologies
must be considered. One promising technology has emerged in the form
of optical computing.
Optical Computers
Optical computers permit the manipulation of every element of an
image at the same time. This parallel processing technique involves
many additions and multiplications occurring simultaneously. Most
digital processors must perform one operation at a time. Even though
the digital processors are very fast, the number of total operations
required to recognize patterns in an image is very large. Using
optical Fourier transformers, an optical processor can operate on the
image and its Fourier transform simultaneously. This permits many
standard image processing techniques, such as spatial filtering and
correlation, to be performed at tremendous rates.
The Fourier transform is formed optically by use of a lens. The
usual case that is considered in optical computing is when the
illuminating source is located at infinity (by use of an auxiliary
collimating lens) and the image transparency is located at a distance
equal to focal length from the transforming lens. The distribution in
the output plane located a focal length behind the transforming lens
is the exact Fourier transform of the input distribution. The Fourier