AN AUTOMATIC UNSUPERVISED PATTERN RECOGNITION APPROACH
Tudor Barbu · 2006
In this paper, we propose an automatic unsupervised classification technique. The method works successfully for any kind of feature vectors, therefore we insist on classification step of recognition process only. First we propose an semiautomatic unsupervised classification approach, based on region-growing method. Then, by eliminating the condition of knowing the number of classes, we obtain an automatic clustering procedure. This method can be successfully applied in various domains which use pattern recognition. Thus, it can be used to perform classification of the components of a large media database, where the number of classes cannot be set through interactivity. Key words: pattern recognition; feature vectors; unsupervised classification; automatic clustering; region-growing, distance, object, media entities. 1.