Page 445 - Introduction to Information Optics

430 7. Pattern Recognition with Optics

(a)

7.29 A target training set is provided in Fig. 7.64(a).
(a) Using a simulated annealing algorithm, synthesize a bipolar compos-
ite filter (SABCF).
(b) Evaluate (by computer simulation) the normalized correlation peak
intensity (NCPI) as a function of training targets.
(c) Plot the NCPI and the discrimination ratio (DR) as a function of
training targets.
7.30 Repeat the preceding exercise for a 16 x 16 pixel frame, and compare
your result with the 8x 8 pixel frame.
7.31 If the antitarget set of Fig. 7.64(b) is included for the synthesis,
(a) Evaluate the SABCF.
(b) Plot the NCPI and DR as a function of training targets.
(c) In view of the preceding results, comment on the performance
capacity for using both target and antitarget sets.
7.32 Refer to the target and antitarget sets of Figs. 7.64(a) and (b).
(a) Synthesize a 3-level SA composite filter.
(b) Plot the NCPI and DR as a function of training targets.
(c) Compare part (b) with the results obtained in the preceding exercise.
7.33 To demonstrate the shift-invariant property of the JTC-NNC in Sec. 7.6,
we have shifted the input "2" upward and left by 8 x 8 pixel, is shown in
Fig. 7.65. Assume that the SLM has 640 x 480 pixels, and each letter is
fitted within a 32 x 32 pixel window.
(a) Show (by computer simulation) the output correlation as obtained
from a binarized JTPS.
(b) What would be the allowable shift of "2" without causing any
possible classified error?
7.34 To investigate the noise performance of the JTC-NNC of Sec. 7.6, assume
that the set of input characters in the preceding exercise is embedded in