Joint Crypto-Stego Scheme for Enhanced Image Protection With Nearest-Centroid Clustering

Abstract

Owing to the exceptional growth of information exchange over open communication channels within the public Internet, confidential transmission of information has become a vital current concern for organizations and individuals. In the proposed content-protection scheme, the decryption key is embedded in the encrypted image by utilizing machine learning, nearest-centroid clustering classifier, followed by least significant bit matching (LSB-M) in the spatial domain. An image is first encrypted with the advanced encryption standard (AES) algorithm in output feedback mode, after which the AES key is embedded into the encrypted image. Preliminary nearest-centroid clustering followed by shuffling the sequence of pixels within the clusters before applying LSB-M makes any attack more complex, as the bits of the key are further dispersed within the encrypted image. In terms of contributions, one contribution is the direct implementation of the proposed security mechanism on color images rather than first converting them into gray tones. Another contribution of the Crypto-Stego method is that, it requires no separate key distribution mechanism to decrypt the information. In addition, a parallel-processing approach is implemented to improve the execution time and the efficiency of the scheme by exploiting system resources. Extensive experiments were performed on RGB images with different resolutions and sizes to confirm the effectiveness of the scheme. The high structural similarity index score confirmed that the overall carrier image and stego-image were unaltered by processing. While an average value over the test images of 0.0594 for mean squared error confirmed that malicious individuals cannot detect the presence of stego data in the cover image. Moreover, negligible pixel intensity histogram changes also validated the effectiveness of the proposed scheme. An average 77% efficiency and 1.5 times speed-up factor were achieved through parallel processing showed the effectiveness of the joint Crypto-Stego method for image confidentiality.

Keywords

• Cryptography
• image processing
• nearest-centroid clustering
• LSB-M
• steganography
• parallel processing