Page 435 - Introduction to Information Optics

420 7. Pattern Recognition with Optics

( a ) ( b )
U,, U I 2 U,N Ikl

U 2I U 22 U 2N Ik22 ik2N

U N, U N2 U NN kN!
Fig. 7.55. Display formats, (a) For input pattern, (b) For the IWM.

A block diagram of the polychromatic neural network (PNN) algorithm is
illustrated in Fig. 7.57. A set of reference color patterns is stored in the NN;
then each pattern is decomposed into three primary color patterns, which are
used as the basic training sets. For the learning phase, three primary color
IWMs should be independently constructed, allowing a multicolor IWM to be
displayed on LCTV2. If a color pattern is fed into LCTV1, the polychromatic
iterative equation is

U lk(n
B
(7.78)

To demonstrate the PNN operation, we used a heteroassociation polychro-
matic training set, as shown in Fig. 7.58. By implementing the heteroassoci-
ation model, as described in Sec. 7.8.1, a multicolor IWM is converged in the
network. When a red color A is presented to the PNN, a corresponding red
Chinese character is translated at the output end. Similarly, if a blue Japanese
Katakana is presented, then a blue color A will be observed. Thus, the PNN
can indeed exploit the spectral content for pattern recognition.

0000

Fig. 7.56. Pixel structure of the color LCTV.