Dual Color Images Watermarking Algorithms Based On Quaternion Transform And Machine Learning

Internet technology has undoubtedly revolutionized our lives with its convenience,while it has also brought numerous security threats,namely the digital product has been copied,tampered and forged illegally.Watermarking is a technology for protecting copyright and information security,and the watermark is invisibly embedded into digital carriers,where video,audio,image and text are involved.Compared with gray images,color images are used more frequently in everyday life.Color cover images and color watermarks are studied in this thesis.Following extensive analysis of the current color image watermarking schemes,and considering that the quaternion matrix could take advantage of the correlation between three color channels of a color image,two robust dual color images watermarking schemes are designed based on machine learning technologies.The primary research work is as follows:A robust dual color images watermarking scheme is put forward based on quaternion discrete fractional angular transform(QDFr AT)and genetic algorithm.To guarantee watermark security,the original color watermark image is encrypted with a4 D Lorenz hyperchaotic system.A pure quaternion matrix is acquired by performing the discrete wavelet transform(DWT),the block division and the discrete cosine transform on the original color cover image.The quaternion matrix is operated by the QDFr AT to improve the robustness of the color watermarking algorithm with the optimal transform angle,which is obtained by a genetic algorithm.To further enhance the scheme’s stability,the singular value matrix of the cover image obtained by the quaternion singular value decomposition(QSVD)is regarded as the embedding area of the watermark.The encryption watermark is also processed by DWT and QSVD.Afterward,the encryption watermark is embedded into the host image’s singular value matrix with the optimal scaling factor to complete the embedding process.Moreover,the value to balance imperceptibility and robustness is optimized with a genetic algorithm.It is shown that the proposed color image watermarking scheme performs well in imperceptibility,security,robustness and embedding capacity.By combining quaternion fractional-order orthogonal Fourier-Mellin moments(QFr OOFMMs)with least squares support vector regression,a dual color images watermarking algorithm is presented based on geometric correction.The color watermark is preprocessed by the 3D sine map to enhance security.The pure quaternion matrix is obtained by operating block division and discrete cosine transform on the low-frequency sub-bands,which are acquired by performing the lifting wavelet transform on the original color host image.The robustness of the watermarking scheme is enhanced by executing the quaternion discrete Fourier transform on the quaternion matrix.Subsequently,the singular value matrix of the watermarked image with embedding operation is yielded by performing the QSVD on the encryption watermark and the processed host image.To better describe the features of a color image,the QFr OOFMMs are defined by generalizing the fractional-order orthogonal Fourier-Mellin moments to the quaternion domain.Before extracting the watermark,geometric distortion parameters to realize geometric correction are estimated by the trained least squares support vector regression based on the low-order QFr OOFMMs.Experimental results illustrate that the designed dual color images watermarking algorithm performs well in defending against both common attacks and geometric attacks.