Stegano-Purge: An integer wavelet transformation based adaptive universal image sterilizer for steganography removal

Abstract

Sterilization is the method of annihilating steganographic data hidden in a cover media. This paper proposes a visual quality preserving image sterilization method that effectively modifies stego-data embedded by various steganography algorithms on random greyscale images. The method applies one-level Integer Wavelet Transform (IWT) on a suspected stego image to produce HH, HL, LH and LL sub-bands. The value of a sub-band coefficient is modified according to its agitation quotient. The coefficient agitation is computed using the proposed metric Unnormalized Hellinger Distance (UHD) between a specific sub-band coefficient and its neighbours. To mitigate the issue of variable noise tolerance of IWT sub-bands, the proposed method performs parameter-controlled adaptive modification. Purification of up to 3 bits is done in the most noise-resilient HH channel and up to 2 bits in moderately noise-sensitive HL and LH channels. The highly noise-susceptible LL channel is sterilized by adaptive bit-flipping. A parameter control algorithm balances the amount of channel-wise bit modification to optimize the visual quality. The stego images for evaluation have been prepared by applying eleven state-of-the-art steganography methods including UNIWARD and HUGO on benchmark datasets of USC-SIPI and BossBase 1.01. The proposed method achieves improved performance in comparison with the state-of-the-art sterilization techniques in terms of the visual quality of sterilized image, embedding scheme-wise percentage of sterilization and the sterilization time.