Improvement And Implementation Of Blind Source Separation Image Encryption

As application of digital image is highly popular, image security has been attracting more and more attention and becoming an important theme of information security. Image encryption is an efficient method for ensuring image security. Although traditional text encryption approaches can be used to encrypt image, they are inefficient since without using characteristics of image data. Considering that images are insensitive to small distortion, blind source separation (BSS) based image encryption utilizes the apparent intractability of the underdetermined BSS problem to encrypt multiple images. Because the key images are independent, and have the same size and number as plain-images, the BSS encryption has very high security.However, the BSS based image encryption has two limitations. Firstly, the order of decrypted images may be different from that of the original images and the values of pixel in some decrypted images may be reversed. Secondly, when encrypting high correlated images, preprocessing must be performed before the BSS decryption. To solve the two problems, this thesis examines the characteristics of BSS based image encryption, proposes two approaches, feature extracting and information embedding, to eliminate the BSS ambiguity, and presents a novel decryption method based on correlation calculation.Feature extracting approach includes direct feature extracting and indirect feature extracting. Before encrypting, the approach extracts features of each plain-image. After BSS decryption, the BSS indeterminacy can be removed by using the features. Information embedding approach utilizes digital watermarking technique to embed a little amount of information. The indeterminacy can then be removed by detecting if decrypted images include the embedded information. Besides, the thesis proposes a novel method for image decryption based on correlation calculation since the key images are known and independent of the original images. Given the encryption equation, correlation matrix between the encrypted images and the key images can be calculated. By ignoring the smaller correlation matrix between the uncorrelated signals, the undetermined encryption matrix is first estimated, and the decrypted images are then obtained. The new decryption method is immune from the BSS indeterminacy and the inseparability of high correlated. Extensive computer simulations and performance evaluation show the efficacy of the proposed methods.