| Similarity measures are the basis of most of the machine learning and pattern recognition algorithms. The choice of the similarity determines the effectiveness of the algorithm in solving the specific problem. This is why finding a relevant similarity measure is an active area of research in machine learning and pattern recognition.; Hamming distance is a simple and efficient similarity measure, but because it was designed to deal with binary vectors, it can not be applied to many problems that uses real-valued vectors. This thesis build upon and extends a generalization of the Hamming distance, Fuzzy Hamming distance, that can operate on real-valued vectors and maintain the same meaning as the Hamming distance: the number of different elements.; To assess the effectiveness of this new measure, FHD is employed in several experiments as basis for a Content Image Retrieval system, a banknote validation system and into a conceptual spaces based, knowledge discovery system. |
