DCT-SVM-based technique for off-line signature verification

Abstract

In this paper, a new method for off-line signature verification is proposed based on discrete cosine transform (DCT). The proposed approach has two stages, namely feature extraction and representation of signature using DCT followed by classification through support vector machine (SVM). The training signature samples are subjected to preprocessing to obtain binarized image, and DCT is employed on the binarized image. The upper-left corner block of size m X n is chosen as a representative feature vector for each trained signature sample. These small feature vector blocks are fed as an input to the SVM for training purpose. The SVM is used as a verification tool and trained with different number of training samples including genuine, skilled, and random forgeries. The proposed approach produces excellent results on the standard signature databases, namely CEDAR, GPDS-160, and MUKOS-a Kannada signature database. In order to demonstrate the superiority of the proposed approach, comparative analysis is provided with many of the standard approaches. © 2014 Springer India.