Hashing-based remote sensing image tamper detection system

Abstract

Considering the large scale of remote sensing images in practical application scenarios, this paper proposes an image detection system by extracting feature hashing of remote sensing images. The feature hashing is generated from different perspectives such as spatial and frequency domain, local and global. For the edge features, an optimized Canny operator based on four pairs of threshold (OCO-4) is employed to calculate the spatial coordinates of rich edge blocks and low edge blocks. In the frequency domain, compression sensing (CS) is exploited to compress and sample Discrete Cosine Transform (DCT) blocks coefficients. For the global texture features, contrast, correlation, energy and homogeneity are extracted by gray-level co-occurrence matrix (GLCM). Furthermore, in order to enhance the security during cloud transmission, pseudo-random numbers are used as secret keys to construct the secure hashes. Consequently, the feature hashes are obtained in both the frequency domain and the space domain, which ensures the effectiveness and uniqueness of feature hashing. Simulation experiments are carried out on four different remote sensing data sets. The results show that the proposed detection system has high sensitivity and a desirable balance between robustness and discrimination, and the average detection accuracy can reach 99.96%.