High-precision privacy-protected image retrieval based on multi-feature fusion

With the production of large amounts of image data in industrial applications, people outsource image computation and storage to the cloud to reduce the storage burden. To prevent privacy disclosure, it is necessary to encrypt images before storage and transmission. However, this encryption operation will greatly limit the utilization of the image, such as image retrieval. Some image retrieval schemes based on single local feature cannot fully reflect an image, and has low retrieval precision. In addition, image retrieval schemes that extract global features need to train a model. If the queried image dataset is different from the training set, the retrieval precision will be low. To improve the retrieval precision, a high-precision privacy-protected image retrieval scheme is proposed. In this scheme, we apply the SM4 algorithm and secure kNN algorithm to encrypt images and features, guaranteeing the security of the data. The construct of the Bkd index tree and the query trapdoor improve the retrieval efficiency. In addition, the feature extraction and the feature fusion technique are adopted to realize the high-precision encrypted image retrieval. Finally, experimental results confirm that our scheme outperforms existing methods in terms of retrieval precision, as well as the speed of image retrieval, trapdoor generation, and index generation.