| 摘要: | 指紋具有勝任的獨特性水準,因為各種特徵可以在每個個體中形成不同的模式
。它是各種多重的驗證要求,例如手機,銀行帳戶,出勤等。 然後,將處理這些資訊
以生成更真實,更準確的資料。 除了指紋識別可以提高安全性外,該方案還容易受到
感測器級別的攻擊。研究表明, 通過使用複製良好的合成手指(如明膠,乳膠,eco
flex,playdoh,木膠等)可以欺騙各種指紋掃描器,這些材料都是基於濕度的,大多數
指紋掃描儀可以可視化,保持性能的預防措施是活度檢測。 提出活體檢測來識別這種
欺騙性吸引,以提高指紋識別系統的安全性。活體檢測是一種功能,用於確定所呈現
的生物特徵樣本是否來自活體。 因此,我們深入利用手工製作的工藝來實現足夠的性
能。我們使用局部二進位模式和相位量化特徵對像素鄰域分佈中的空間和頻域進行共
軛。同時,在預處理中,我們使用圖像轉換來製作更多的變化圖像。為了封裝雜訊的
可能性,我們添加了小波變換作為雜訊去除。最後,我們使用一種突出的機器學習方
式映射學習階段,即、支援向量機 (SVM)。我們的實驗以準確性和平均錯誤率進行
評估。所提出的方法在 LivDet 2011,LivDet 2013 和 LivDet 2015 上的平均錯誤率降低
4.2,2.1 和 5.1 方面取得了可持續的結果。;Fingerprint has a competent level of uniqueness because various features can form a
different pattern in each individual. It is a verification requirement in various multiple, such as
mobile phones, banking accounts, attendance, etc. This information will then be processed to
generate more factual, accurate data. Besides fingerprint recognition can improve security, the
scheme turns out to be vulnerable to attacks at the sensor level. Studies have shown that it is
possible to trick various fingerprint scanners by using well duplicated synthetic fingers such
as gelatin, latex, eco flex, playdoh, wood glue, etc. These materials are humidity-based, and
most fingerprint scanners can visualize the preventive measures in maintaining the
performance is liveness detection. Liveness detection is proposed to identify this kind of
spoof attracts to improve security for the fingerprint recognition system. Liveness detection is
a function that determines whether the presented biometric sample originated from a live
body. Thus, we deep exploited the handcrafted process to achieve adequate performance. We
conjugate the spatial and frequency domain in pixel neighborhood distribution using local
binary pattern and phase quantization feature. Meanwhile, in preprocessing, we use image
translation to make more variation images. And to encapsulate the noise possibility, we added
the wavelet transform as the noise removal. Finally, we map the learning stage using a
prominent machine learning way, i.e., support vector machine (SVM). Our experiment is
evaluated with accuracy and average error rate. The proposed method has achieved
sustainable results in terms of reduction in average error rates 4.2, 2.1, and 5.1 on LivDet
2011, LivDet 2013, and LivDet 2015. |
