Self-organizing cortical processing with visual feature selection for pattern recognition

Pattern recognition has been studied extensively, and many algorithms have been established. It generally makes use of discriminant functions to learn the pattern in data. These discrimant functions are developed to be simplistic so as to warrant fast computations. In addition, simple evaluation functions are easier to learn because there are lesser parameters to estimate. However, this simplicity may not work well when new ‘pattern’ in data surfaces. Humans recognize an object or pattern from surrounding world in split second; however this involves many processing in the human visual system. Human gathers most of the sensory information through sight. Visual-perceptual processing covers approximately one-fourth of the cortex. Visual information processing is also the most complex, most studied sensory system of the brain. It is envisaged that if the visual cortex can process information in such a lightning speed, there should exist some combinations of feature selection and pattern classification which are close enough to provide such capability. The motivation behind the research of this thesis is to establish a computational framework that attempts to emulate the visual cortical processing in the human brain. The aim is to recognize a pattern in short computation time even when sparse data is presented.DOCTOR OF PHILOSOPHY (SCE