A new blind attack procedure for dct-based image encryption with spectrum learning

It is an important problem nowadays to ensure the security of encrypted multimedia contents on the Internet. In this paper, a new blind attack procedure called NZCA is proposed to decrypt images that are encrypted by DCT-based methods. It threatens almost all existing DCT-based image encryption methods without the need of knowing the encryption algorithm details. The parameters of the NZCA can be adaptively selected via unsupervised learning with spectrum analysis of image projection. The effectiveness of the NZCA has been quantitatively evaluated by comparing the decrypted images with the edge sketches of original images. Our experiments on a number of real-world images show that NZCA is very powerful. Furthermore, we point out that a special form of inter-block shuffle, full inter-block shuffle (FIBS), is immune to the NZCA. Hence we suggest that all DCT-based image encryption methods for JPEG, MPEG, H.26x, etc. employ the FIBS so as to enhance the security.