Deep learning and chaotic map based encryption algorithm for color face image

Abstract Currently, a large number of applications rely on face recognition technology to ensure the proper functioning of their services, such as mobile payment, employee attendance tracking, and identity authentication. The protection of user facial privacy data from leakage has become critically important. To address this issue, a deep learning and chaotic map based encryption algorithm for color face image (i.e., DCEA) is presented. Firstly, YOLOv8 detection model is taken to locate facial regions in the plain face image. Subsequently, a novel chaotic map combined with an improved two-dimensional cellular automaton is used to generate pseudorandom sequences, which are then applied to perform row-column scrambling and XOR encryption exclusively on the detected facial regions, yielding a partially encrypted face image. Experimental results demonstrate that DCEA can yield an information entropy of 7.9999 and maintain extremely low pixel correlation. Furthermore, the region of interest (ROI)-based strategy achieves a speedup ratio of approximately ten times for high-resolution images compared with full-image encryption schemes. Fortunately, DCEA can also effectively balance high-level security with computational efficiency, making it suitable for real-time privacy protection.