Machine Recognition in Complex Domain - High Dimensional Neurocomputing: Growth, Appraisal and Applications - page 129

Chapter 7
Machine Recognition in Complex Domain
Abstract Machine recognition has drawn considerable interest and attention from
researches in intelligent system design and computer vision communities over the
recent past. Understandably there are a large number of commercial, law enforce-
ment, control and forensic applications to this. We human beings have natural ability
to recognize persons at a glance. Motivated by our remarkable ability, a series of
attempts [ 1 - 4 ] have been made to simulate this ability in machines. The develop-
ment of human recognition system in machines is quite difficult because the natural
objects are complex, multidimensional, and corresponds to environmental changes
[ 3 , 5 - 7 ]. There are two important issues that need to be addressed in machine recog-
nition: (1) how the features are adopted to represent an object under environmental
changes and (2) how we classify an object image based on a chosen representation.
Over the years, researches have developed a number of methods for feature extraction
and classification. All of these, however, have their own merits and demerits. Most
of the work is related to the real domain. The outperformance of complex-valued
neuron over conventional neuron has been well established in previous chapters.
Few researchers have recently tried multivariate statistical techniques in the complex
domain, like complex principal component analysis (PCA) for 2D vector field analy-
sis [ 8 ] and complex independent component analysis (ICA) for performing source
separation on functional magnetic resonance imaging data [ 9 , 10 ]. But, no attempts
have been made to develop techniques for feature extraction using their concepts.
This chapter presents formal procedures for feature extraction using unsupervised
learning techniques in complex domain. Efficient learning and better precision in
result offered by feature extractor and classifier, considering simulations in complex
domain, figure out their technical benefits over conventional methods. Notably, the
success of machine recognition is limited by variations in features resulting from the
natural environment. These may be due to instrument distortion, acquisition in an
outdoor environment, different noises, complex background, occlusion and illumi-
nation. A solid set of examples presented in this chapter demonstrate the superiority
of feature representation and classification in complex domain.