K-nearest neighbor classifier for signature verification system

The paper presents an off-line signature verification system using k-nearest neighbor classifier. Global features from the signatures are extracted using radon transform. For each registered user in the system database a number of reference signatures are enrolled and aligned for statistics information extraction about his signature. Extreme points warping algorithm is used to align two signatures. During the k-nearest neighbor classifier training, a number of genuine and forged signatures are chosen. A test signature's verification is established by first aligning it with each reference signature for the claimed user. The signature is then classified as genuine or forgery, according to the alignment scores which are normalized by reference statistics, using standard pattern classification techniques. Using a database of 2250 signatures (genuine signatures and skilled forgeries) from 75 writers in the proposed signature verification system a performance of approximately 80% is achieved.