Signature Recognition using VFI5 Algorithm and Wavelet Preprocess.

Abstract

Biometric is a science in recognizing the identity of a person based on their physical or behavioral traits. Today, one of the most widely applied biometric object is hand-written signature. In this research, we tried to identify offline handwritten signature using VFI5 (Voting Feature Intervals 5) classification algorithm. The VFI5 classification algorithm has been used previously on non-image objects with promising result. The feature used in the classification is the grayscale value of the image and the signature image dimension used in this research is 40×60 pixels which means originally there are 2400 features that relatively large to compute. Thus, Haar wavelet transform is used to reduce the dimension. This is done by taking only the approximation image of the transformation result, and leave the horizontal, vertical, diagonal details. The wavelet decomposition is applied until the fifth level, creating new low level images with 20×30, 10×15, 5×8, 3×4, 2×2 dimension respectively for the first, second, third, fourth, and fifth decomposition levels. The approximation images then classified using VFI5 algorithm, with 97.5%, 95%, 90%, 65%, and 62.5% accuracy respectively for the first, second, third, fourth, and fifth decomposition levels. It turned out that VFI5 algorithm can be used to identify images with wavelet transform to reduce the image dimension with good result, and the higher decomposition level applied, the more accuracy dropped. Keywords: signature identification, wavelet image dimension reduction, image classification, VFI5 algorithm.