Research On Digital Multimedia Passive Forensics Based On Similarity Analysis

With the progress of digital multimedia technology, the multimedia equipments, such as digital cameras and digital video recorders, have become popular. Meanwhile, a lot of video editing software with powerful function, such as Photoshop and Mokey, has been widely used, so nonprofessional people are able to tamper with an image or a video without difficulties. These images or videos having been tampered usually cover up the truth and even make the "eye-stickers" look realistic. If these tampered images or videos are used in the news, scientific discovery and judicial forensics, it will be sure to bring serious influence on society. So it is becoming increasingly important and urgent to authenticate the veracity and the integrity of videos, which has already become the research hotspot in recent years.Image and video are the most common medium in our life, and the multimedia forensics has become the hot spot of academic study. Because of the complexity of video, such as large data volumes, high-dimensional and non-linearity, compared with image forensics, video forensics is much more difficult. Hence, it is becoming particularly important to reduce the data quantity, the detection complexity and the computation time. Aiming at the multimedia forensics, our main work includes the following two aspects:(1) Splicing image tamper detection based on color temperature similarity analysis. In image processing, splicing tampering of image usually will destroy the illumination consistency of natural image. As a manifestation of illumination consistency, color temperature is especially suitable for the detection of splicing image. According to the principle of optical imaging, due to the changes in lighting conditions, there will be a certain color temperature difference between different images in the process of photographic imaging. Based on it, a splicing image tamper detection based on color temperature similarity analysis is proposed. The main idea of the algorithm is:using the technology of target detection, the method in question pinpointed the faces in the image, and then estimated the color temperature of all those faces and the reference color temperature. After calculating the discrepancy between human face color temperature and reference color temperature, the method found out the detection target with abnormal color temperature discrepancy as the tampering target, and then located the precise position of the tampering target. The empirical study showed that the algorithm is effective.(2) Aiming at the duplication of selected frames from a video to another location in the same video, we proposed an effective similarity-analysis-based method for frame duplication detection. The method is implemented in two stages. In the first stage, the features of each frame are obtained via SVD (Singular Value Decomposition). Next, the Euclidean distance is calculated between features of each frame and the reference frame. After dividing the video sequence into overlapping sub-sequences, the similarities between the sub-sequences are calculated, and then our algorithm identifies those video sequences with high similarity as candidate duplications. In the second stage, the candidate duplications are confirmed through random block matching. The experimental results show that our algorithm provides detection accuracy that is higher than the previous algorithms, and it has an outstanding performance in terms of time efficiency.The innovative points of this paper mainly include:(1) Aiming at the forgery of splicing image, our algorithm combines face detection technology with image tamper detection, realizing the automatic positioning of detecting targets in the images. Our algorithm only calculates the color temperature of the target area to judge whether the image has been tampered and locates the juggled position, which greatly reduce the amount of calculation.(2) Aiming at the forgery of frame duplication in the same video, our algorithm can be applied to videos in any format which has better adaptability, whereas some methods were intended only for MPEG videos. Additionally, the detection accuracy with this algorithm is better than that of the previous algorithms. Specifically, our algorithm exhibits an outstanding performance with respect to time efficiency.