簽名辨識之研究

Abstract

在這篇論文中我們利用改良的鍊碼(modified chain codes)及所提出新 的運算法則運用於離線(off-line)的簽名辨識上 我們用複寫紙來偵測人 們簽名時的力道 首先，我們找出簽名的上下輪廓(upper profile and lower profile) 這些輪廓(profiles)可以幫助我們將這個簽名分類，但 尚未決定真假 在特徵抽取上，我們結合灰階變化和鍊碼(chaincodes)的 特性，產生固定維度(dimension)的正規向量(normalized vector) 最後 我們用類神經網路，指導式模糊適應Hamming網路(supervised fuzzy adaptive Hamming network)，來判斷真假 在模擬方面，我們收集了34 個人的真跡和偽造簽名，結果顯示有良好的辨識率 We propose a novel off-line algorithm and modified chain codes for recognizing signatures. We use a carbon paper to detect force distributionswhile people write their signatures. First, we find the signature contour andgenerate upper and lower profiles. These profiles help us to classify the givensignatures. We also combine gray-scale and chain codes characteristics to extract structure and force distribution features, which are then transformedinto normalized vectors. We then use a supervised fuzzy adaptive Hammingnetwork (SFAHN) to interpret the feature vectors in order to determine whetherthe signatures are genuine or not. simulation results show the proposed algorithm has a good recognition rate.