Image understanding based on independent component analysis (ICA) and stochastic learning automata (SLA)

In analyzing data, we often extract new information including features from the existing and original data for the purpose or reducing the dimension of feature space and achieving an improved performance. This dissertation looked at a new paradigm of signal and information processing combined. The first step (in this dissertation) taught us that a combined energy/power based algorithm plus information/memory based algorithm is an attractive and practical method to deal with the ever-increasing system complexity. The overall contribution of this research was to extent the current capability of independent component analysis (ICA) along with the stochastic learning automaton (SLA), which can produce a number of features that do not carry unwanted features for object recognition, to make the process simpler and, as a result faster data retrieval. ICA tries to solve the blind source separation problem in which sensor signals are unknown mixtures of unknown source signals.; We propose a new standard algorithm that combines independent component analysis (ICA) and stochastic learning automata (SLA) that can produce a number of features that do not carry unwanted features for object recognition. Independent Component Analysis (ICA) is emerging as the new standard area of signal processing and data analysis. ICA tries to solve the blind source separation problem in which sensor signals are unknown mixtures of unknown source signals.; On the other hand, stochastic learning automaton is a system, which modifies its control strategy on the basis of its experience in order to reach optimization performances in spite of unpredictable changes in the environment where it operates. According to literature survey, both Independent Component Analysis and stochastic learning automata have attracted considerable interest due to their potential usefulness in a variety of engineering problems that are characterized by high dimensionality, non-linearity, and high level of uncertainty. One of the inherent difficulties in all forms of object recognition is that the input image cannot be completely recognized and the automatic characterization of the contents of the image is not completely known for recognition. We will experiment with different images from different fields.