The use of hyperspectral images as cover data in information hiding

Abstract

The paper tells the features of using hyperspectral images (HSI) as a container for embedding hidden information. It is proposed to use the accuracy of the classification of HSI (as the most common problem solved with the help of HSI) to control the permissible level of introduced distortions instead of the indicator of visual imperceptibility. We need to increase the volume or robustness of embedded information while maintaining acceptable classification quality. For this it is proposed to change the embedding parameters for different HSI channels in proportion to their significance in classification. As an example, three fundamentally different methods of embedding information are considered: the method based on quantization index modulation(QIM), the method based on spread spectrum and the method based on interpolation. Experiments have shown the absolute advantage of the proposed approach over the standard one (parameter values constant for each channel) for all three methods. Experiments have confirmed that the proposed approach makes it possible to fill a significant part of the volume of the HSI (about 3/4 for the method based on QIM) almost without deterioration in the quality of classification. And for the interpolation-based method, it is possible not only not to worsen, but to improve the quality of classification along with the embedding of hidden information.