Research On Image Authentication Algorithms Based On Perceptual Hashing

The development of digital image processing techniques has brought many newchallenges to image security. As such, image information protection has become one ofthe most important research problems. There exist many image authentication techniquesthat could solve the problem effectively. Traditional image authentication and encryptiontechniques are usually based on the bit stream of an image file. However, their identifi-cation degrades seriously in the process of transmission and storage. Therefore, imageauthentication techniques based on perceptual hashing are proposed.The perceptual hashing models expand the traditional models, diversify and refinethe granularity of the image authentication. Although the perceptual models outperformthe traditional bit stream models, they still have two main limitations. First, there areinadequate perceptual features to support the models; second, the security is not wellconsidered along with the high robustness.To overcome the above limitations, we firstly made a detailed survey on existing per-ceptual hashing algorithms, based on which, we analyzed the requirements of algorithmdesign. Then we proposed a novel framework of image authentication and designed twonew perceptual hashing algorithms with regarding to different patterns and granularities.Compared with previous work, we have made the following three contributions:First, we proposed an algorithm based on contourlet transform. In this algorithm,we defined two new concepts"fast rotation registration scheme"and"semi-diffusion".The fast rotation registration scheme turns the rotated image into the original style whilesemi-diffusion combines the diffusion and non-diffusion to improve the security of imageauthentication. To the best of our knowledge, we are the first to use contourlet transfor-mation for perceptual feature extraction.Second, we proposed an algorithm based on structure skeleton. We took the struc-ture skeleton of contourlet features as the perceptual information. The structure skeletoncould identify the content changes of the images, which makes our image authenticationscheme resistant to tampering. The tamper localization is achieved by processing theinformation of the perceptual features.Finally, we investigated the existing security standards of image authentication based on perceptual hashing, based on which we defined our security standard. Thenwe analyzed our algorithms against existing work according to the security standard.