Novel Methods For Encrypting Images And Speeches Based On Blind Source Separation

With the developement of information society, information has become the most important and valuable resources to be transmitted conveniently and promptly through Internet. However, the globality and openness of Internet severely threaten the information security. At the same time, image and speech communications become more and more widely used as the multimedia technology is progressing rapidly. Therefore, the security of images and speeches has become a worldwide problem.Usually, regular data encryption methods can be used to encrypt images and speeches. To further enhance their security level, new theories and new methods need to be explored to give new schemes for encrypting images and speeches. Blind source separation (BSS) is one active branch of stochastic signal processing. It can recover source signals from their observed mixtures without knowing the source signals and the mixing coefficients. BSS has received considerable attention in recent years, and has been applied to many fields such as wireless communications, biomedical engineering. However, there have been only a few reports for its application to signal encryption thus far.Motivated by the fact that the security of many cryptosystems relies on the apparent intractability of the computational problem such as the integer factorization problem, this dissertation explores the feasibility of BSS-based signal encryption, and exploits the intractability of the underdetermined BSS problem to present novel encryption methods for encrypting images and speeches. The BSS-based encryption method can have the perfect key characteristics of one-time pads cipher, which is unconditionally secure. If the key signals pseudorandomly generated are completely random, the proposed method is approximately unconditionally secure. Extensive computer simulations and performance analyses demonstrate the efficiency of the BSS-based encryption.The main contributions of this dissertation are as follows:(1) By studying the BSS theory and the cryptography, the underdetermined BSS problem has been totally exploited to establish a novel method for encrypting images with high security. Two necessary conditions for generating the key signals are given based on BSS theory and the key characteristics of one-time pads cipher, and their necessity is shown. Two mixing matrixes for underdetermined encryption are constructed by using the L-row decomposable theorem, and their inseparability is also proved.(2) By using some waveform information about the original speech, a novel method for encrypting speeches is proposed based on BSS. The proposed method encrypts the original speech frame by frame, and each frame is divided into multiple segments for underdetermined mixing encryption. Due to the indeterminacy of BSS, the BSS decrypted segments may have permutation and scale ambiguities compared with the original segments. Considering that the BSS indeterminacy is caused by lack of prior information about sources, the numbers of zero-crossings, the maximum and the minimum of each original segmen are recorded and used to eliminate ambiguity of the BSS decrypted speech in permutation and amplitude. As a result, the decrypted speech signal has been recovered with very high quality.(3) Since the computational load of the BSS algorithms usually has nonlinear relation (e.g., square, cubic, etc) with the number of the source signals, a fast method for decrypting images is given by using the knowledge of the key images on the receiving side, and by employing the technique of adaptive noise cancellation to half reduce the number of the source signals for the BSS decryption. Besides, strongly correlated images are well decrypted with the existing ICA algorithms by decorrelation of the encrypted images before BSS.(4) Based on the masking technique, image cryptosystems with dual encryption are established by using BSS to add another encryption level, which includes "specific mixing" and "BSS", besides the existing encryption methods.(5) Several different BSS adaptive algorithms are unified to one form with various nonlinear functions through reasonable derivation and conversion. The stability criterion for choosing nonlinear function in the unified form is described. The combined criterion for choosing the nonlinear function that gives attention to several desired performances such as separation performance, robustness and convergence speed is given.In summary, this dissertation proposes a novel application of BSS to image encryption and speech encryption. The BSS-based encryption methods can be largely used for secure transmission of images and speeches through Internet.