Pixels Who Violate Our Privacy! Deep Learning for Identifying Images’ Key Pixels

In recent years, face recognition as an effective biometric technique has received significant attention. The worldwide application of face recognition, however, raises strong concerns regarding the individuals’ privacy. This problem has been particularly addressed by encrypting all pixels of an image, which is computationally heavy. In this paper, we propose a new framework using deep neural network that determines the most identifiable pixels (instead of all pixels) of an image for encryption. Our experimental result shows that our proposed approach significantly reduces the number of pixels required to be encrypted.