AAU-Net: Attention-Based Asymmetric U-Net for Subject-Sensitive Hashing of Remote Sensing Images

Abstract

The prerequisite for the use of remote sensing images is that their security must be guaranteed. As a special subset of perceptual hashing, subject-sensitive hashing overcomes the shortcomings of the existing perceptual hashing that cannot distinguish between “subject-related tampering” and “subject-unrelated tampering” of remote sensing images. However, the existing subject-sensitive hashing still has a large deficiency in robustness. In this paper, we propose a novel attention-based asymmetric U-Net (AAU-Net) for the subject-sensitive hashing of remote sensing (RS) images. Our AAU-Net demonstrates obvious asymmetric structure characteristics, which is important to improve the robustness of features by combining the attention mechanism and the characteristics of subject-sensitive hashing. On the basis of AAU-Net, a subject-sensitive hashing algorithm is developed to integrate the features of various bands of RS images. Our experimental results show that our AAU-Net-based subject-sensitive hashing algorithm is more robust than the existing deep learning models such as Attention U-Net and MUM-Net, and its tampering sensitivity remains at the same level as that of Attention U-Net and MUM-Net.

SDTU-Net: Stepwise-Drop and Transformer-Based U-Net for Subject-Sensitive Hashing of HRRS Images

“…Compared with cryptographic hash (such as MD5 and SHA1), digital signature and blockchain, perceptual hashing is able to map images with the same content into the same hash sequences. It can be used to realize image authentication [12], image retrieval [13] and image copy detection [14].…”

“…In [11], MUM-Net is used for subject-sensitive feature extraction. In [12], AAU-Net is used to implement critical feature extraction tasks. In fact, other deep learning networks such as U-net [22], M-net [23], Attention U-net [24], MultiResUNet [25] and Attention ResU-Net [26] all can be used for subject-sensitive feature extraction.…”

“…The difficulty of deep learning based subject-sensitive hash algorithm lies in how to balance robustness of the algorithm with tampering sensitivity [12], and dropout technology can improve this difficulty to some extent.…”

“…The experimental data in this paper includes two categories: datasets used to train deep neural networks, and datasets used to test subject-sensitive hash algorithm. SDTUnet is trained using the same training dataset in [11] and [12], which is a dataset based on WHU building dataset [38] and contains 3166 pairs of training samples. Datasets testing the performance of subject-sensitive hashing algorithms will be presented in Section IV-C.…”

“…2 (b) and (c) is subjectunrelated and subject-related tampering, respectively. Subject-sensitive hashing, derived from perceptual hashing, overcomes the above shortcomings [11]- [12] and can realize subject-sensitive authentication. It can perform stricter authentication on the main ground object information used by users.…”