On pattern description by languages

This paper starts with a brief review of the syntactic (linguistic) approach to pattern recognition. One-dimensional and high dimensional languages for pattern description and recognition are domenstated by application examples. Several problems motivated by practical applications of syntactic pattern recognition are discussed. In applying syntactic approach to pattern recognition problems, abstract pattern primitives are often required. A sentence structure representing a pattern can then be analyzed with respect to the given pattern grammers. Pattern matching or/and syntax analysis (parsing) is in general used as a recognition procedure. In practical applications, noise and distortion involving in the pattern measurements are unavoidable. A perfect recognition of primitives and the relations among them would be very difficult, even with excessive proprocessing and sophisticated recognition techniques. Consequently, ambiguity often occurs. Also, at the present, patterns grammars are primarily constructed by the designer. The quality of a heuristically constructed grammar in terms of its pattern descriptive capability is difficult to evaluate. Such a grammer may generate sentences representing unwanted patterns or it may not be able to describe the patterns in the same class which have not occurred before. There is a need to have systematic learning or inference procedures which will learn (with generalization) a pattern grammer from a set of training patterns. With these practical considerations in mind the following topics and their potential applications are then discussed: (1) transformational grammers, (2) stochastic grammers and languages, (3) syntax analysis in the presence of errors and (4) grammatical inference.