Development of Efficient Color Image Encryption Techniques Using Evolutionary Approaches

Abstract

The advancements in multimedia applications are rapidly increasing nowadays. A number of confidential images are transferred over the public networks day-by-day. Therefore, secure transmission of images has become a significant research area. Several image encryption techniques have been designed to achieve the particular necessities raised by different users. The image encryption techniques convert confidential image into a noisy image using secret key. The actual image is recovered if and only if the receiver has an authenticated secret key. Various techniques such as chaotic maps, evolutionary, DNA, cellular automata, etc. have been used to encrypt the images. Image encryption differs from text encryption as a result of few numbers of inherent characteristics such as huge size and redundancy. The traditional text encryption techniques fail to handle huge size and redundancy of images. The main challenges of image encryption are robustness against attacks, keyspace, key sensitivity, and diffusion. The chaotic maps are extensively utilized in the field of image encryption to generate the secret keys. However, these maps suffer from parameter tuning problem. Recent studies have shown that the improper selection of parameter values makes secret keys generated from chaotic system vulnerable. Therefore, many meta-heuristic techniques have been introduced in the literature for image encryption to improve the selection of chaotic system’s parameters. However, these techniques suffer from poor computational speed. Also, designing an efficient multi-objective fitness function is still a challenging issue. To overcome these issues, meta-heuristic based image encryption techniques are proposed in this thesis. An efficient image encryption technique in nonsubsampled contourlet transform using genetic algorithm (IGN) is proposed. Initially, nonsubsampled contourlet transform is utilized to decompose the input image into sub-bands. The beta chaotic map is used to develop a pseudo-random key that encrypts the coefficients of sub-bands. A multi-objective fitness function is designed to find the optimal parameters for beta chaotic map. Inverse nonsubsampled contourlet transform is performed to obtain an encrypted image. The performance of IGN is compared with well-known meta-heuristic based image encryption techniques. Experimental results reveal that IGN provides better computational speed and high encryption intensity. However, genetic algorithm may suffers from local optima and premature convergence issues. To deal with these issues, an image encryption technique using differential evolution in nonsubsampled contourlet transform (so called IDN) is proposed. In this technique, two new concepts are utilized to encrypt the images in an efficient manner. The first one is Arnold transform, which is used to permute the pixel position of an input image to generate a scrambled image. The second one is differential evolution, which is used to tune the parameters required by a beta chaotic map. The entropy of an encrypted image is used as a fitness function. IDN is compared with seven well-known image encryption techniques over five well-known benchmark test images. The experimental results reveal that IDN outperforms the existing techniques in terms of security and better visual quality. To improve the local search ability of IDN, the initial conditions of an intertwining logistic map are generated by utilizing memetic differential evolution and Arnold transform (IIMA). It utilizes local chaotic search to improve the search ability of standard differential evolution. Initially, the color image is decomposed into red, green, and blue channels. Arnold transform is used to shuffle the pixel position of all three channels to develop the shuffled channels. Afterward, memetic differential evolution is used to optimize the parameters required by intertwining logistic map. The correlation coefficient and entropy are used as a fitness function. The intertwining map generates the secret keys to encrypt the shuffled color channels. The encrypted color channels are combined to obtain the encrypted image. Extensive experiments are carried out by considering IIMA and the existing competitive image encryption techniques. Experimental results reveal that IIMA provides higher efficiency and security as compared to the existing image encryption techniques. To secure the secret key, an efficient Image encryption technique based on Secure hash algorithm (SHA-3), adaptive differential evolution, and Lorenz-like chaotic system known as ISAL is designed. ISAL utilizes SHA-3 along with adaptive differential evolution to reduce the issues associated with Lorenz-like chaotic system. In ISAL, adaptive differential evolution is used to optimize the input parameters for Lorenz-like chaotic system. SHA-3 is used to generate a secret key based on the input image. The optimized parameters and external secret keys are used to generate the initial values for Lorenz-like chaotic system that make it sensitive to an input image and provide resistance against both known-plaintext and known-ciphertext attacks. ISAL is compared with five well-known image encryption techniques over four color benchmark test images. The experimental results reveal that ISAL outperforms the existing techniques in terms of security and quality measures. The noise and enhancement attacks are also applied to test the robustness of ISAL. Although, IGN, IDN, IIMA, and ISAL provide better encryption results than the existing techniques. But, they suffer from poor computational speed especially in case of highresolution images. Therefore, an Image encryption technique based on Fourier-Mellin moments and intertwining logistic map (IFIM) is proposed. It uses multi-objective nondominated sorting genetic algorithm based on reinforcement learning (MNSGA-RL) to optimize the required parameters of intertwining logistic map. Fourier-Mellin moments are used to make the secret keys more secure. Thereafter, permutation and diffusion operations are carried out on the input image using secret keys. The performance of IFIM is evaluated on five well-known benchmark images and also compared with seven well-known existing encryption techniques. The experimental results reveal that IFIM outperforms the others in terms of entropy, correlation analysis, unified average changing intensity, and number of pixel change rate. The simulation results reveal that IFIM provides a high level of security and robustness against various types of attacks. To improve the computational speed, IFIM is implemented in a parallel fashion using master-slave environment. The run time analysis has been done to determine the computationally expensive operations. Thereafter, IFIM operations are divided into master and slave jobs. Message passing interface (MPI) is used for intercommunication between master and slave nodes. The simulation results show that parallel IFIM provides a significant improvement in computational speed as compared to the existing techniques and sequential IFIM. The proposed techniques (i.e., IGN, IDN, IIMA, ISAL, and IFIM) satisfy the various statistical parameters and offer tangible resistance to differential, occlusion, and chosen plaintext attacks on gray as well as color images. The proposed techniques achieve not only a desirable level of security, but, also high efficiency and better computational speed. Therefore, the designed techniques are well suitable for real-life applications.