A privacy-preserving scheme for JPEG image retrieval based on deep learning

With the great increase of digital images, content-based image retrieval (CBIR) has been proposed to facilitate the usage of images. Local users need to outsource CBIR to cloud services, which leads to privacy leakage. Based on deep learning, this paper proposes a retrieval scheme of encrypted JPEG images. In this scheme, the image owner sends the corresponding encrypted image to the cloud server. The cloud server extracts the Variable-length Integer (VLI) code length from the encrypted image and uses Convolution Neural Network (CNN) to learn a value of fixed dimensions as the feature vector. We have shown through experiments that our scheme is safe, with good performance and retrieval accuracy.