Abstract:
This thesis is concerned about the detection or pattern recognition of unsegmented
targets under varying illumination with the application of normalized joint transform
correlation technique. When the illumination within a scene cannot be controlled, the
outputs for similar targets can be quite different. Pattern recognition techniques usually
use the correlation operation to detect targets by means of a threshold on the correlation
plane. But as correlation peak height is proportional to target intensity, high intensity
non-targets can cause false alarms and dark objects can also be missed. A number of
methods have been so far introduced for detection of targets under transformation of
intensity. Phase only and synthetic discriminant functions have been proposed for
normalization which is realized in the frequency domain. In this thesis work, an
efficient implementation of normalized correlation in the space domain has been
proposed to achieve real time discrimination between intensity varying similar nontargets.
The previous method like complementary reference joint transform correlator
was proposed which is useful in recognizing only binary images. But this scheme works
almost successfully for binary and gray level images. Some researchers also pointed out
that the morphological correlation is intensity invariant only when the object is brighter
than the reference. But the proposed scheme has better processing speed as it utilizes
linear correlation than that of morphological correlation, which is slower in
computation. In this work, a high performance optical correlator has been developed
which gives almost equal correlation peaks for all the targets, whether the targets are
brighter or darker than the reference and whether there is noise in the input scene or not.
To achieve higher discrimination ration between dissimilar non-target objects a post
processing technique has also been introduced. Computer simulation shows satisfactory
performance of the proposed scheme in getting intensity invariant pattern recognition.
