Experiments in Bayesian and neural pattern recognition with applications to textured-image classification
D.M. Rose, Aly A. Farag · Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE · 1993
ABSTRACT In this paper, two approaches to pattern recognition will be applied to textured image classification. The first approach is statistical, based on the Bayesian description, and the second approach isneural. I. INTRODUCTION Pattern recognition is based in human perception and cognition. Human existence takes the formof patterns, whether it is the formation of language, the process of speech, the drawing of pictures, or theunderstanding of images. Our perceptive powers are well adapted to such pattern—processing tasks, despitemajor variations, distortions, or omissions in a pattern. We would like to implement similar capabilities inmachines, enabling them to understand what we say, to read what we write, and in general, to respond in intuitively understandable ways. In other words, we would like to build into our machines the same pattern—information processing capabilities that we possess, resulting in machines that are easier to use andmore efficient in handling real world tasks.Traditional pattern recognition concentrates on the mathematical or computer—science aspectsof pattern—fonnatted information processing with emphasis on statistical pattern recognition. There is alsoconsiderable interest in the use of mathematical syntactic structures and the use of fuzzy logic to provide a