An Enhanced Statistical Approach to Identifying Photorealistic Images
IWDW '09 Proceedings of the 8th International Workshop on Digital Watermarking
ICIP'09 Proceedings of the 16th IEEE international conference on Image processing
Distinguishing computer graphics from photographic images using local binary patterns
IWDW'12 Proceedings of the 11th international conference on Digital Forensics and Watermaking
| Hi-index | 0.00 |
The advancement of digital techniques makes it much easier than ever before to tamper with digital media such as digital image and video. The tampered image or video comes to appear so perceptually realistic that the trustworthiness of digital image or video cannot be taken for granted any more. Because it is not common in practice that the authentication data such as digital signature is embedded into a medium when the media is generated, the passive-blind techniques to authenticate the digital image or video are important for the applications such as criminal investigation, news reporting, and intelligence analysis. In this dissertation, the passive-blind techniques in the following three areas have been investigated: (1) image splicing detection; (2) computer graphics identification; and (3) double MPEG video compression detection. The proposed techniques have been implemented in the two-class pattern recognition framework. Image splicing is the process of making a composite picture by cutting and joining one or multiple regions from the same or different photographs without post-processing. To differentiate the authentic images from spliced images, the distinguishing image features are extracted by exploiting both magnitude and phase information of a given image. The first part of image features is the statistical moments of characteristic functions of the test image, its prediction-error image, and their wavelet subbands. Furthermore, the approximation (LL) subband at different levels is individually erased, and inverse wavelet transform is applied in order to enhance high frequency components. From these reconstructed images with the LL subbands erased, the moments of wavelet characteristic functions form the second part of features. Finally, the statistics of 2-D phase congruency array associated with the above-mentioned reconstructed images are the third part of features for splicing detection. Due to the high photorealism achieved by computer rendering software, the computer graphics can be considered as a convincing form of image forgery nowadays. To distinguish computer graphics from photographic images, the statistical moments of characteristic function of the images and their wavelet subbands are used as the distinguishing features. The influence of image color representation systems on the feature effectiveness is investigated. Furthermore, to remove the redundant or unnecessary features, the binary genetic algorithm is applied to select the most useful features. The video is often tampered in spatial domain by altering video content, dropping, inserting or reordering the frames. For MPEG video file, the tampered video often needs to be re-encoded and re-saved, in which double compression occurs. By identifying the traces introduced by double MPEG compression, the possible tampering operation can be revealed. The problem of double MPEG compression detection is approached by analyzing the probability distribution of the first digits of the non-zero MPEG AC coefficients. To determine whether the MPEG video is doubly compressed, a decision rule using GOP as the detection unit is proposed.