Zero Watermarking Algorithm for Hyperspectral Remote Sensing Images Considering Spectral and Spatial Features

Most existing zero-watermarking algorithms for remote sensing images are designed for panchromatic or multispectral data. When applied to hyperspectral data, these methods fail to fully utilize the unique characteristics of hyperspectral images, resulting in poor robustness. This study proposes a novel zero-watermarking method specifically tailored for hyperspectral remote sensing images, leveraging both spectral and spatial features. First, principal component analysis (PCA) is applied to reduce the dimensionality of the hyperspectral image, and the leading principal component is used for K-means clustering. The frequency of each category is calculated based on the clustering results, and the category frequency vector is used as the spectral feature vector. Simultaneously, Zernike moments are employed to extract spatial features, generating the spatial feature vector. These spectral and spatial feature vectors are combined to form the carrier image feature vector. Next, the watermark image is scrambled using Zigzag scanning and binarized. The MLNCML chaotic system generates a random sequence, which is XORed with the binarized watermark sequence to produce the final binary watermark sequence. Finally, the carrier image feature vector and the binary watermark sequence are XORed to generate the zero watermark. The experimental results show that under the common attacks such as geometric attack, noise attack, filtering attack and combination attack, the NC values of the proposed algorithm are high, all above 0.9, the values of BER are low, and the NC values of fusing spectral and spatial features are higher than that of extracting only spectral or spatial features. These results show that the proposed method effectively exploits the data characteristics of hyperspectral remote sensing images, is robust to various attacks, does not cause loss of accuracy, and is suitable for copyright protection of hyperspectral remote sensing image data.