High-Resolution Remote Sensing Image Integrity Authentication Method Considering Both Global and Local Features

Abstract

Content integrity of high-resolution remote sensing (HRRS) images is the premise of its usability. Existing HRRS image integrity authentication methods are mostly binary decision-making processes, which cannot provide a further interpretable information (e.g., tamper localization, tamper type determination). Due to this reason, a robust HRRS images integrity authentication algorithm using perceptual hashing technology considering both global and local features is proposed in this paper. It extracts global features by the efficient recognition ability of Zernike moments to texture information. Meanwhile, Features from Accelerated Segment Test (FAST) key points are applied to local features construction and tamper localization. By applying the concept of multi-feature combination to the integrity authentication of HRRS images, the authentication process is more convincing in comparison to existing algorithms. Furthermore, an interpretable authentication result can be given. The experimental results show that the algorithm proposed in this paper is highly robust to the content retention operation, has a strong sensitivity to the content changing operations, and the result of tampering localization is more precise comparing with existing algorithms.

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

“…The information regarding various bands of RS images often has certain differences, but the existing perceptual hash algorithms (including subject-sensitive hashing) for RS images often do not take this difference into account. The existing perceptual hash algorithms for RS images [13][14][15][16] generally perform a grayscale operation on the original image in the preprocessing stage and then extract the characteristics of the obtained grayscale image. This obviously does not take into account the band differences of RS images.…”

“…Perceptual hash [11,12] overcomes the shortcomings of cryptography technologies and digital watermarking technology and has strong application prospects. Some scholars have carried out related research on perceptual hash authentication algorithms of RS images, leading to some encouraging results [13][14][15].…”

Semi-U-Net: A Lightweight Deep Neural Network for Subject-Sensitive Hashing of HRRS Images

“…Perceptual hash originated from digital watermarking technology, in which it is used as embedded watermarking information, and later became an independent technology. Perceptual hash has been widely used in image retrieval [25][26][31][32], image copy detection [33][34], and image integrity authentication [20][21][22][23][35][36].…”

“…Our subject-sensitive hashing algorithm based on Semi-U-Net follows the general steps of subject-sensitive hashing, which consists of preprocessing, subject-sensitive feature extraction, compression encoding, and encryption, as shown in Figure 4: 1) Preprocessing mainly performs operations such as resampling and channel fusion on HRRS images, so that HRRS images can meet the input requirement of Semi-U-Net. If the HRRS image is too large in practical applications, the method similar to [21][22] [29] can be used to divide the HRRS image into grids first, and then our algorithm is used. Since our research focuses on lightweight deep neural networks, we do not consider that case.…”

“…The design of the perceptual hash algorithms for HRRS images should be based on its data characteristics while taking into account the application environment. At present, the research of perceptual hash of remote sensing images mainly includes the perceptual hash algorithm for HRRS images [20][21][22] and the perceptual hash algorithm of multispectral remote sensing images [23].…”