Content Based Image Authentication Algorithm Robust To Geometric Attacks

With the rapid development and the wide application of multimedia technology and network technology, storage, processing and distribution of digital multimedia products become easier and easier. However, this also leads to potential security problems. Consequently, authenticating the authenticity and the integrity of digital multimedia content is becoming increasingly pressing. Digital images have a large amount of redundant information, and they can often be compressed by different compression standards, and represented by different formats. However, digital image is usually subject to various signal processing manipulations, such as JPEG compression and geometric transform during distribution, which may maintain the visual quality of the original image. Hence, the content based image authentication that can tolerate normal signal processing and be fragile to malicious tamper has become a hot spot, as it meets practical requirements.Image authentication technique is considered as an important way to ensure the security of image storage and transmission. This paper analyzes the drawbacks of the existing image authentication methods, and then studies the content based image authentication schemes resilient to geometric attacks. The main work of this thesis is summarized as follows.First, an image content authentication algorithm based on Zernike moments and image normalization is proposed. This method takes advantage of the semi-fragile characteristics of Zernike moments to distinguish malicious attack from common signal processing. Firstly, image normalization is used to transform original image into standard image, and then Zernike moments of the low frequency subband of DWT image are computed. Finally, the image hash, deemed as the watermark in this paper, is generated from the robust features that are determined by the optimal moment set. When detecting, the normalized Hamming distance is employed to weigh the similarity of the image hash by comparing the image hash and the extracted watermark. Experimental results show that the proposed scheme is not only robust against common signal processing and geometric attacks such as rotation, scaling, etc., but also sensitive to malicious content modification manipulations made to the image data.Second, a content authentication scheme, which is robust to geometric transformations, is proposed by using the time-frequency localization property of DWT and the geometric property of Radon transform. Firstly, the robust features are first extracted via performing the discrete wavelet transform, followed by the Radon transform. Then, the extracted features are taken as the watermark, embedded into the original image. In the detection stage, radon transformation and image moment are utilized to detect and invert geometric transformations, and then content based image authentication can be carried out by extracting the watermark. Experimental results show that the proposed method is robust against content-preserving modifications and is capable of detecting malicious tampering and locating the tampered area as well.