Smart Analytical Signature Verification For DSP Applications

TL;DR

This paper presents a biometric signature verification system using wavelet and cosine transforms combined with SVM classification, achieving high accuracy on benchmark datasets.

Contribution

It introduces a novel feature extraction method combining DWT and DCT for signature verification and applies dimension reduction for improved performance.

Findings

Achieved an EER of 8.7% on benchmark datasets.

Attained an average verification accuracy of 91.3%.

Demonstrated effectiveness of combined wavelet and cosine transforms.

Abstract

Signature verification is an authentication technique that considers handwritten signature as a biometric. From a biometric perspective this project made use of automatic means through an integration of intelligent algorithms to perform signal enhancement function such as filtering and smoothing for optimization in conventional biometric systems. A handwritten signature is a 1D Daubechies wavelet signal (db4) that utilizes Discrete Wavelet Transform (DWT) and Discrete Cosine Transform (DCT) collectively to create a feature dataset with d-dimensional space. In the proposed work the statistical features characteristics are extracted from each particular signature per data source. Two databases called Signature Verification Competition (SVC) 2004 database and SUBCORPUS 100 MCYT Bimodal database are used to cooperate with the design algorithm. Furthermore dimension reduction technique is applied to the large feature vectors. A system model is trained and evaluated using the support vector machine (SVM) classifier algorithm. Hence an equal error rate (EER) of 8.7 percent and an average correct verification rate of 91.3 percent