Feature Selection for Forensic Handwriting Identification

Current paper describes the use of a feature selection technique to reduce the number of features while the goodness set is selected on a framework for forensic handwriting identification. A sequential forward search and an evaluation criterion based on dependency were used to obtain a goodness subset (GS) to improve the identification rate. The accuracy of the system applied to 100 different writers and taking account all features (N = 81) is 58%, whereas the accuracy based on goodness subset (GS) is 80% applied to the same number of writers. The validation of results was verified initially against all the features and later against some empirically set of features. Results are comparable to others in the literature on graphometric features.