Picture description language for syntactic pattern recognition : of seismic patterns and English characters
In this study, a syntactic pattern recognition system is proposed to recognize seismic patterns and English characters. The system includes two parts: analysis, and recognition. The analysis part consists of preprocessing, picture primitive selection and clustering analysis. Two thinning techniques are used in preprocessing. A formal picture description schema-picture description language is used as the basis for picture processing system to generate pattern representation. Non-supervised clustering technique is used in characters recognition and supervised clustering technique is used in seismic pattern recognition. The former uses a set of input figures to generate pattern classes through PDL and clustering analysis, whereas the latter uses known figures as training patterns. The clustering techniques are mainly on the pattern-to-pattern basis. There are four clustering techniques used, as K-nearest neighbor classification rule, minimum-spanning tree, algorithm a, and algorithm [beta]. An input sentence (a pattern) is compared with sentences in a formed cluster, one by one, or with the representation of the cluster by computing the Levenshtein distances to determine the similarity. A set of English characters is used to illustrate the system. They are 51 characters from nine different classes: D, F, H, K, P, U, V, X, and Y. The analyzed 2-D synthetic seismic patterns are bright spot, pinch-out, flat spot, gradual sealevel fall, and gradual sealevel rise pattern.