Outsourced Secure Face Recognition Based on CKKS Homomorphic Encryption in Cloud Computing

Abstract

With the enhancement of the performance of cloud servers, face recognition applications are becoming more and more popular, but it also has some security problems, such as user privacy data leakage. This article proposes a face recognition scheme based on homomorphic encryption in cloud environment. The article first uses the MTCNN algorithm to detect face and correct the data and extracts the face feature vector through the FaceNet algorithm. Then, the article encrypts the facial features with the CKKS homomorphic encryption scheme and builds a database of the encrypted facial feature in the cloud server. The process of face recognition is as follows: calculate the distance between the encrypted feature vectors and the maximum value of the ciphertext result, decrypt it, and compare the threshold to determine whether it is a person. The experimental results show that when the scheme is based on the LFW data set, the threshold is 1.1236, and the recognition accuracy in the ciphertext is 94.8837%, which proves the reliability of the proposed scheme.

Response Time of Cloud-Based Facial Recognition System Utilizing Homomorphic Encryption

“…Factors significantly influencing the RTT of the Euclidean distance acquisition process included the computational cost of the distance acquisition process, which increases in proportion to the number of registrants, and the substantial increase in the data size of the facial features due to homomorphic encryption [9]. The average size of plaintext facial features was 4,160 bytes, whereas the average size of homomorphic ciphertext was approximately 334,300 bytes.…”

“…However, homomorphic encryption has its own set of complexities. The latency for facial recognition employing homomorphic encryption significantly exceeds that in the case of plaintext usage [8,9]. Consequently, fulfilling the required response time for practical use may be challenging [10,11].…”