The Research Of Forged Image Detection Aiming At Splicing And Double JPEG 2000 Compression

The widespread use of the imaging equipments and a variety of image editing software make the phenomenon of digital image tampering happen frequently. According to this, it is of extremely significant to detect whether the given image is authentic or not. The traditional active digital image detection technology has poor usefulness and versatility because it depends on pre-embedded signature or watermark information. Therefore, the passive digital image authentication technology, which does not require any prior conditions, has become a novel and pretty important research direction in the area of information security. As a result, this dissertation focuses on the passive digital image authentication. Since there are various tampering technologies, the corresponding authentication methods are different and complex. This dissertation specializes in the authentication algorithm of the spliced image and doubly compressed JPEG2000 image, which are the typical ones in the natural image tampering detection.Splicing is a common scheme of image tampering. After splicing, the texture characteristics of an image and the relationship between its pixels have changed. As long as these artifacts were extracted, they can be used as the features to discriminate between the authentic image and the spliced image without any pre-embedded information, which is known as blind detection of the image. Through analyzing the texture characteristic, this dissertation proposes a run length histogram feature for the detection of the spliced image. Moreover, the dissertation combines the run length histogram feature with the Markov feature and proposes a specific implementation solution for the detection of the spliced image. At last, the experimental results were given to demonstrate that the detection accuracy of the integrated features can reach to 89.89%, which outperforms the detection accuracy of the algorithm only uses the run length histogram feature or the Markov feature.For the passive detection of the JPEG2000 image, this dissertation proposes a blind detection algorithm for the JPEG2000 image by using the Markov feature in space domain. First, the dissertation analyzes the changes of the relationships between pixels of the singly compressed JPEG2000 image and doubly compressed JPEG2000 image. Then, this dissertation makes a comparison of the Markov features in DCT domain and in space domain. Based on this theory, the dissertation proposes the feature extraction method by using the Markov matrices in spacial domain. The thresholded prediction-error images are generated by subtracting horizontal, vertical, main diagonal and minor diagonal pixel values from current pixel values respectively. Then the prediction-error images were thresholded with a predefined threshold T in order to remove the big values which reflect the difference of the image content other than the artifacts of the JPEG2000 compression. At last, the dissertation uses the Markov random process to model the thresholded prediction-error image in order to utilize the second-order statistics. Support vector machine is used as the classifier. Experimental results have shown that the algorithm has achieved good performance for most bit rates and the proposed approach has outperformed the prior arts.