- 學位論文
指紋辨識科技在犯罪鑑識應用之研究
The Research on the Forensic Applications of Fingerprint Identification Technology
摘要
刑事犯罪現場上經常發現在非吸水性的紅包袋上有模糊不清楚、不完整或重疊的指紋影像，第一線的鑑識人員不易採集且容易造成誤判。本研究首先在指紋前處理時採用化學藥劑-氰基丙烯酸酯(Cyanoacrylate)煙燻法來採集指紋，減少採集指紋的瑕疵；其次應用人工智慧兩種深度學習模型架構U-net和全卷積網路(Fully Convolutional Network，FCN)，針對所採集到模糊或重疊指紋來驗證影像分割和影像重建。實驗結果顯示，在深度5和6的U-net模型的均方誤差為0.0060至0.0061，非常低。同時，U-net模型深度5和6的平均絕對誤差從0.0172到0.0168也很低，U-net模型的重建結果優於其他方法。 本研究主要是結合鑑識科學、人工智慧科技及管理方法等跨領域合作，所開發的模式除減少鑑識人員比對指紋的時間，使指紋鑑定速度更迅速、縮小比對範圍及降低鑑識人員的生理負荷，提高指紋辨識的效率，避免發生人為誤判。 安全管理是目前在各領域中非常重要課題，在工業場域內始終會面臨安全漏洞的持續威脅。尤其在生產製造現場內的原物料容易遭遇竊取，常造成企業的嚴重損失。如企業能提升門禁安全系統的指紋辨識效率，將能確保其重要原物料受到更完善的保護、有效降低財產損失、並提升安全管理的效能。
並列摘要
Unclear, partial or overlapping fingerprint images are often found on non-absorbent red envelopes at the scene of criminal crimes, which is difficult for frontline forensic personnel to collect and easily lead to misjudgment. This study first utilizes the cyanoacrylate fuming method as a chemical reagent for fingerprint preprocessing, aiming to reduce defects in fingerprint collection. Subsequently, two artificial intelligence deep learning models, U-net and Fully Convolutional Network (FCN), are applied to verify image segmentation and reconstruction of the collected unclear or overlapping fingerprints. Experimental results reveal that the mean squared error (MSE) of the U-net models with depths of 5 and 6 is extremely low, ranging from 0.0060 to 0.0061. Similarly, the mean absolute error (MAE) of U-net models with depths of 5 and 6 is also very low, ranging from 0.0172 to 0.0168. The reconstruction results achieved by the U-net model surpass those of other methods. This study integrates forensic science, artificial intelligence technology, and management methodologies through interdisciplinary collaboration. The developed approach not only reduces the time required for fingerprint matching by forensic personnel but also accelerates fingerprint identification, narrows the scope of comparison, alleviates the physiological burden on forensic staff, improves fingerprint recognition efficiency, and minimizes the risk of human error in judgment. Security management is a very important topic in various fields at present, and it will always face the persistent threat of security loopholes in industrial fields. Raw materials in manufacturing sites are easy to be stolen, which often causes serious losses to the enterprises. If the fingerprint identification efficiency of the access control security system can be improved, it will ensure that its important raw materials are better secured, effectively reduce the property losses and efficiently improve the security management of the enterprises.