Emerging Trends in Deep Learning-Based Image Encryption: A Survey

With the exponential growth of digital communication and data exchange, image security has emerged as a critical concern across domains such as healthcare, military, social media, and surveillance. With this surge, ensuring the authenticity and security of these images has become a critical concern for researchers. Traditional encryption methods, while widely used, face limitations when applied to the growing volume and complexity of image data. As a result, the use of deep learning (DL) models for image encryption has gained significant attention due to their adaptability and advanced capabilities. This paper offers a detailed review of recent developments in image encryption that leverage DL approaches. First, we present our motivations on the use of deep learning to secure images and the mutual advantages that arise for both the researchers in this field and the final users. Afterward, we investigate several state-of-the-art DL-based encryption approaches, with a summary of their major procedures and the salient characteristics that distinguish each one. In this regard, we present a comparison that unifies the most representative contributions in the state-of-the-art. Then, in comparison with other surveys in the field, we discuss the most representative open challenges and potential future research directions, including the urgency of setting a unified standard of evaluation for the security of DL-based image encryption.