Purpose. In order to improve the effect of image copyright protection and detect whether an image is tampered illegally, we introduce an image copyright protection and tampering detection scheme of ROI (Region of interest) image based on image feature sequence in NROI (Non Region of interest) image. We have evaluated this scheme with some performance measures and the results show it is effective. Methodology. We formulate the scheme using the copyright watermarking and the fragile watermarking. With the deep learning, memristor chaos, Arnold transform and extend zigzag transform, the watermarkings are generated and embedded into ROI image in DCT (Discrete cosine transform) domain using the feature sequence of NROI image. Findings. We first completed the division of ROI and NROI image and get the feature sequence of NROI image using deep learning and memristor chaos. Then by using the sequence and some methods such as Arnold transform, we obtained the scrambling copyright watermarking and the new fragile watermarking of each image grouping and embedded them into ROI image. Originality. We realize the extraction of image feature sequence in NROI image using deep learning and memristor chaos. It is applied to generate and embed the scrambling copyright watermarking and the new fragile watermarking into ROI image. The research on this aspect has not been found at present. Practical value. We have completed some validation experiments with some performance measures. The results show it can completely satisfy the need of secure transmission. This scheme features strong robustness and security.

Purpose. With the growing demands of image processing on the Internet, image compression and encryption have been playing an important role in image protection and transfering. In this paper we will investigate deep learning technology in image compression, and chaotic logistic map in image encryption, to obtain a scheme in image compression and encryption. We have evaluated this scheme with some performance measures and results show it is effective. Methodology. We formulate the scheme using deep learning and chaos. With the deep learning technology, levels of features are extracted from an image and a certain level of features can be used as a compressed representation of the image. Chaos is used to encrypt the compressed image. Findings. We first introduceda five-layer Stacked Auto-Encoder model, which is trained by the Back Propogration method, and then we obtained the compressed representation of an image. By using the logistic map method, a pseudo-stochastic sequence is generated to encrypt the compressed image. Originality. We conducted a study of image compression and encryption. Image characteristics are extracted from an arbitrary level of our deep learning model, and they are used as the compressed representation of the image. The research on this aspect has not been found at present. Practical value. We have evaluated this scheme on several randomly selected images. And results show it is robust and can be widely used for most images.