簽名確認系統( SVS )是一種以簽名為生物特徵來確認簽名者身份的生物認證系統。目前已被廣泛的應用在許多需要確認使用者身份的場合中。簽名是生物特徵的一種，不像密碼( password )有容易遺忘、被盜用或是被猜測的危險性。並且比起其他生物認證系統，由於以簽名來確認身份的方式由來已久，因此更容易被大眾所接受。 本論文將實做一個線上中文簽名確認系統，此系統由全域性特徵與區域性特徵兩個獨立的模組組成簽名確認核心，全域性模組在特徵抽取階段選擇了目前文獻中一些經常被利用到的特徵，區域性模組則提出了一個新的方法—抽取簽名的資料串列( data string )。不管是全域性或區域性模組，都會以筆劃外型 (實筆) 與提筆軌跡 (虛筆) 作為抽取特徵之對象。特徵抽取結束之後，將得到全域模組特徵集與區域模組特徵集，接著分別對兩特徵集作真偽簽名之區別力分析，依特徵之區別力高低對其重要性作排名，接著再選取前幾名的特徵形成個人化的特徵集。在簽名確認的階段，當測試簽名與參考樣本進行特徵比對時，將會依照特徵之重要性而給予不同的權重。實驗結果顯示新的資料串列區域特徵是相當有效的簽名特徵，也驗證本系統有不錯的能力，來分辨真實簽名與模仿簽名。 Signature verification is a kind of biometrics verification which can verify the identity of the signer using signature feature as the biometrics feature. Currently, it has already been applied to a lot of occasions needing the confirming of user identity. It is not easy to forget your own signature or forge other person’s signature like password. Comparing with other biometric systems, it is admitted as the most popular way which can be accepted by everybody in identity verification. In this thesis, an online Chinese signature verification system is devised to achieve the goal of identity verification. The proposed signature verification system is composed of two main modules including global and local modules. Almost all previous literatures emphasize on global module by choosing the global features of signatures. In addition to global module, local module is also proposed by extracting a novel set of features that is the data strings of signatures. No matter global or local module, the stroke trajectory (real-stroke) and the pen-up trajectory (virtual-stroke) will always be used as the target for feature extraction. After performing the feature extraction process, global and local feature sets will be obtained. Then, the two feature sets are analyzed carefully and the importance of the each feature is ranked depending on its power in distinguishing the genuine signature from forgeries. Next, the topper features are chosen to form the personalized feature set. In signature verification process, different weights are assigned to different features when conducting feature matching between the test signature and reference templates so that optimal performance can be obtained. Experimental results demonstrate that the proposed system is effective in signature verification.