Validation of Handwriting Individuality Using Distance Measures and Dichotomies Validation of Handwriting Individuality Using Distance Measures and Dichotomies

In classiication problems such as writer, face, nger print or speaker identiication, the number of classes is very large or unspeciied. To establish the inherent distinctness of the classes, i.e., validate individuality, we transform the many class problem into a dichotomy by using a \distance" between two samples of the same class and those of two diierent classes. A measure of conndence is associated with individuality. We deal with following issues: i) individuality validation: establishing the validity of individuality of classes using the dichotomy model, ii) comparison between polychotomy and dichotomy: comparing polychotomy in feature domain and dichotomy in distance domain from the view point of tractability vs. accuracy, iii) distance measures: use and evaluate several distances, iv) eecient search: nearest-neighbor algorithms for distance measures v) applications: designing and analyzing an algorithm for writer identiication for a known number of writers and a method for handwritten document image indexing and retrieval, and vi) discovery: mining a database consisting of writer data and features obtained from a handwriting sample, statistically representative of the US population, for feature evaluation and to determine similarity of a speciic group of people. Based on conjectures derived from experimental observations, we present theorems comparing the performance of the polychotomizer and dichotomizer. Results comparing newly deened his-togram and string distance measures with conventional measures are given. Applications of the methods to the domain of on-line and oo-line character recognition are addressed.