Information Security And Image Fusion Algorithms Based On Fractional Transforms

With rapid development of the internet and information technology, information hasbrought great convenience to people's lives and also brought many security risks. Thus,the information security technology has become increasingly important. As the mainway to express information, image information security has become a focus of attention.Image encryption and image watermarking are two security technologies. Exploring newalgorithms for image information security technology has significant theoretical and prac-tical value. This work focuses on the image information technology in the transformeddomain. Since the image fusion can be considered as one kind of visible watermarkingand is widely used in remote sensing, this work also studied image fusion algorithms. Theresearch work has been summarized in detail as follows.One color image encryption algorithm has been proposed. The widely used red-green-blue (RGB) color model can not be adapted very well to the color understood byhuman, because the color image is not simply synthesized by three primary-color imagesto one image in practice. The color should be described by intensity, hue and saturation.The intensity-hue-saturation (IHS) color model describes how humans naturally respondto and apperceive color, which satisfies the above discussed very well. The RGB colorimage is converted into IHS space. In the IHS space, three components are encryptedindependently using discrete fractional random transform (DFRNT) and Arnold transform(AT), respectively. AT is a kind of position scrambling. DFRNT is a kind of encryptionwith the secrecy of pixel value and pixel position simultaneously. The randomness ofDFRNT ensures the high security of the proposed algorithm. The fractional order ofDFRNT and iteration numbers of AT are extra encryption keys to enhance the security.Comparing to the classical double-random-phase encoding, this proposed method cansave storage space of the encryption keys required to be stored for decryption, meanwhilewith the higher security. Moreover, the proposed method is robust against typical attacks.The robustness of DFRNT in image watermarking have been investigated and com-pared. DFRNT has excellent mathematical properties inherited from the discrete frac-tional Fourier transform (DFrFT), in addition to the randomness of its own. The robust-ness of watermarking schemes based on DFRNT is superior than those based on dis- crete cosine transform, DFrFT and discrete Fourier transform. Due to the randomnessof DFRNT, the spectrum energy is dispersed uniformly and randomly, which provideshigher capacity while ensuring better robustness against common attacks. DFRNT itselfcan serve as an additional secret key, which ensures the higher security of watermarkingas compared to those based on three other transforms. Moreover, the feature of real outputof DFRNT at half period for real number is convenient for storage and can save storagespace of image data.One optical image watermarking algorithm based on fractional Fourier transform(FrFT) has been proposed. The host image and watermark are placed close to the inputand output planes of FrFT, respectively, then the watermark is embedded using randomphase encoding. The electro-optical hybrid setup is provided for the watermarking imple-mentation in optics. The whole watermarking system can be considered as a packagingsystem. Just inputting a host image, a corresponding watermark can be detected or ex-tracted. There is no distortion to the host image in both the watermark embedding andextracting processes, which ensures the ideal imperceptibility and the original content ofhost image can be completely restored after extracting watermark. This proposed algo-rithm is fast and convenient for real-time implementation. Since the host image is in thephase not in the amplitude, when the strength of distortion is greater than some degree,the extracted watermark tends to be stable, which shows good robustness.One remote sensing image fusion method based on DFRNT has been proposed.The DFRNT is introduced in the image fusion field. The randomness of DFRNT makesthe changed information by fusion random, which introduces lower in?uence than thesame-strength changes at the concentrated location in spectrums. This ensures less spec-tral distortion. The uniformity of DFRNT assures most of the acceptable fusion resultswhen distortions occur in any position of the spectrum, which makes the method withsome robustness. In the DFRNT domain, different fusion rules are adopted in the spec-trum information component and the spatial information component for different goals,respectively. The proposed method is applied to fuse the real multi-spectral TM imageand the SPOT panchromatic image. The fused image is found to effectively preserve boththe spectral information of the multi-spectral image and the spatial information of thepanchromatic image, and preserve more spectral information comparing to the conven-tional fusion methods. One remote sensing image fusion method based on the expectation maximization(EM) algorithm and covariance intersection (CI) approach has been proposed. The imagefusion problem is considered as an estimation problem by which we estimate the idealscene of multi-spectral images at the resolution of panchromatic images. Most estimationmethods assume input sources are independent. In practice, there is correlation betweeninput sources. To this end, the CI which can produce unbiased and consistent estimate forany degree of cross-correlation between the input sources through convex combination,is employed. Since the covariance information of the input source is usually unknown inpractice, the EM is employed to provide a maximum likelihood estimate of the covariancematrix. Therefore, the EM and CI is combined to develop an image fusion method. Incontrast to the conventional fusion methods, the proposed scheme takes cross-correlationamong data sources into account, and thus provides consistent and accurate estimatesthrough convex combinations. Performance evaluation of the proposed method usingreal remote sensing images is accomplished and shows that this method can reach thefusion aim very well, and the spectral information is preserved more effectively than theconventional image fusion methods.One maximum likelihood approach to joint image registration and fusion has beenproposed. Both image registration and fusion are formulated as estimation problems.Instead of estimating the registration parameters and the ideal fusion scene separatelyas in the conventional way, EM approach is employed to joint registration and fusion.At each iteration, fusion result is obtained in the E-step and registration parameters areupdated in the M-step. The fusion performance is used as the criteria to evaluate theregistration accuracy. Hence, registration parameters can be automatically tuned so thatboth fusion and registration can be optimized simultaneously. The mean square error ofestimations is consistent and close to the Cramer-Rao bound.