Performance and accuracy analysis of signature verification using deep learning techniques

Abstract

The biometric system is an accurate, reliable and rugged tools with traditional identification techniques for various applications. A person uses signature verification to avoid automatic entries of forgers. There are two methods in signature verification such as static and dynamic methods, the stored signatures are called static method and online signatures are called dynamic method. The methods of integer wavelet transform and affine discrete wavelet transform are used to identify the signature function. The extracted features are classified into advanced methods to find matching and non-matching handwritten signatures from the test results. newlineSignature verification is widely used for personal verification. Manually verifying handwritten signatures is extremely time consuming and the significant risk of mistakes might threaten property security and possibly social stability. Verification can be done using an online or offline application. In this work signature validation can be based on signal processing. This signal process will process on the scanned signals of offline computer signatures. The proposed handwritten signature verification system consists mainly of three phases: pre-processing, feature extraction, and classification of the output signals. During the first phase of the process, the obtained signal is pre-processed for the reduction of unwanted noise. In second phase, the features like angle, pressure, input vector and sequence of impulses are extracted using Integer Wavelet Transform (IWT) and Affine Discrete Wavelet Transform (ADWT) techniques. Finally in the third phase, the obtained values are evaluated with the proposed Back Propagation Neural Network (BPNN) and Resilient Spectral Neural Network (RSNN) classifier for the identify of genuine and forgery signature. newline