Detection of Fraudulent Alterations in Ball-Point Pen Strokes Using Support Vector Machines

Fraudulent addition to cheques, wills, contracts, and other legal documents may result in serious consequences leading to an irreparable damage in terms of human suffering as well as severe financial loss. The increasing graph of loss due to such a white collar crime is a matter of serious concern. In this paper we propose a mechanism for detection of alteration in ball-point pen strokes using pattern recognition techniques. A large set of features based on color and texture is extracted from images of documents. To find a set of discriminatory features, a neural-network-based feature analysis technique is used. Finally, Support Vector Machine (SVM) is used for the detection. The model selection is done using cross-validation in conjunction with some constraint on false positive rate (FPR) that is demanded by the problem domain. The results are very encouraging.