以機器學習為基礎的加密臉部辨識方法及其在數位監控上之應用

Abstract

臉部辨識是廣為人知的影像識別應用，並且在當代社會很常被應用來做身份的辨別。然而，許多系統卻忽視了保護臉部影像的重要性。若是臉部的影像未被保護起來，惡意人士便能竊取、複製影像，並偽裝成為其他人。為了解決這個問題，我們設計了一個能同時保護臉部影像的臉部辨識的系統。我們使用 DeepID2 卷積類神經網路來提取臉部特徵、使用最大期望演算法來來協助進行臉部辨識。為了確保臉部影像的安全，我們使用同態加密來將臉部特徵加密起來，使最大期望演算法能在密文域中做計算。基於不同的安全性考量，我們針對社區設計了三套不同的門禁系統，並在最後對三個系統做辨識準確率與執行速度等實驗，比較各優缺點。

Face Verification is a well known image analysis application and wildly used for recognizing individuals in contemporary society. However, in real applications, some of recognition systems ignore the importance of protecting the facial images that are used for verification. If the facial images are not protected, malicious people can steal and copy the images to disguise as someone else. To conquer this problem, we design a secure face verification system that can also protect the facial images to be imitated. In our work, we use the DeepID2 convolutional neural network to extract the feature of a facial image and use the EM algorithm to do the facial verification problem. In order to keep the facial images privacy, we use the homomorphic encryption scheme to encrypt the facial data and compute the EM algorithm in the ciphertext domain. Based on difference privacy concerns, we build up three face recognition systems for surveillance or entry and exist control of a local community. Lastly, we conduct experiments of accuracy and time consuming and compare pros can cons between these systems.