On Applying Fast and Efficient Methods in Pattern Recognition

This chapter presents a new approach in pattern recognition which relies on using fast and efficient techniques in the implementation of basic pattern recognition procedures. More specifically, concerning the pattern preprocessing and feature extraction stage, a novel method is employed in the design of the relevant algorithms. Namely, the Image Block Representation methodology is used in the construction of efficient algorithms for quantitatively analyzing and processing patterns, like 2D moment computation for improved feature extraction. It is based on exploiting the properties of simple operators applied to pattern representation and manipulation. Second, concerning the pattern classification stage, a novel methodology is presented for designing improved pattern classifiers based on the concept of effectively training feedforward neural networks using constrained optimization techniques. The two methods, developed in the present research effort as new architectural principles for designing improved pattern recognition systems, are evaluated through extensive experimentation demonstrating reliability and enhanced performance in several real world problems.