Page 408 - Introduction to Information Optics

7.5. Pattern Recognition Using Composite Filtering 393

6 8 10 12 14 16 18
Quantization Level

Fig. 7.32. (a) Detectability and (b) accuracy as a function of quantization level.

thesized. We have subsequently implemented this set of QCFs in a JTC, by
which the detectability and accuracy of detection as a function of quantized
levels are plotted in Figs. 7.32a and b, respectively. We see that both
detectability and accuracy improve as the quantization level increases. How-
ever, they slow down somewhat when the quantization level is higher than 10
levels. We have also plotted the discriminability and the reliability of detections,
as shown in Fig. 7.33. We see that both discriminability and reliability improve
as quantization level increases, but discriminability levels off as N > 10.
Since Fisher ratio (FR) is affected by background noise, we have added 21
random noise levels into the test targets. In other words, the noisy background
is formed by independently adding a zero-mean Gaussian noise to each pixel
element of the test target. The FR for different quantization levels as a function
of noise standard deviation is plotted in Fig. 7.34. Note that target discrimina-