Study On Motion-Adaptive Frame Deletion Detection For Digital Video Forensics

With the development of the network and the popularity of digital cameras, digital visual media data is spreading with an alarming number and speed each day, and becomes one of the most important information carriers in human society. At the same time, with the development of digital media processing algorithms and the maturation of related software, the manipulation of visual media data has become a common phenomenon, which may cause problems in science, politics, judicature, economy, sports and other aspects. Therefore, authenticity of digital visual media has become a key issue of today.Traditional methods for verifying the authenticity of media data comprise digital watermark and digital signature. Such technology uses a proactive way for the identification process, which requires manual watermark or signature embedding while recording the scene or before the distribution of the media data. Therefore put a high demand for data owners and recording devices. So that the range of its applications is limited.Digital visual media forensics is an emerging technology which verifies the authenticity of media data from a passive perspective. Forensic algorithms utilize the traces which had been left by the data acquisition process or the post-processing operation to analyze and restore relevant operations. Such method does not need to embed any additional data, and thus has a great potential for practical application. The research of digital visual media forensics began from about 2000, and developed rapidly in the last ten years. And now it has become a multidisciplinary cross-type research field.Since digital visual media forensics technology is still in the development stage, there are still some problems to be solved. Such as the insufficiency of the study for digital videos, not robust against attack algorithms, unable to cope with complex tampering environments, and so on.According to these problems, the thesis focuses on the study of the forensic method of frame deletion manipulation for videos with variable motion strength, and conducts a systematic study on related issues. The research consists of three parts as the following:1. Motion-adaptive frame deletion forensics. The motion strength of the video contents usually changes over time, which may cause the variation of the characteristics of the inter-frame data. From an inter-frame perspective, we propose a motion-adaptive frame deletion detection method for digital video forensics. First, we analyze the interference frames in complex tampering environments, and distinguish the frame deletion point with its major interference frame according to the fluctuation feature of the frame motion residual. On this basis, we analyze the influence of the variable motion strengths to the fluctuation feature. And propose an intra-encoding elimination algorithm for enhancing the fluctuation feature, so that it can be adaptive to video sequences with variable motion strength.2. The combination algorithm for frame deletion forensics. The advantages and limitations of existing forensic methods are often found to be complementary. The combination of such methods could obtain a significantly improved detection results. This part analyzes the limitations of the original frame deletion detection algorithm, and then presents a complimentary algorithm proposed in literature. On the basis of analyzing the principles of their complementarity, we integrate the two algorithms through a certain way, and aim at extending the application range of the original algorithm while enhancing its detection performance.3. Forensic tools for frame deletion forensics. Due to the imperfection of the media data and the complexity of the tampering process, single forensic algorithm often fails to provide sufficient credibility for evidence a manipulation. The application of the forensic tool set is an effective way to solve such a problem, but relative research is still rare. To this end, the thesis propose two forensic tools that are based on automatic color equalization and mean gradient, respectively. And use them as a post-processing for the frame deletion detection algorithm for enhancing the credibility of the forensic procedure in complex environments. We also explore the application of the proposed forensics tools in counter anti-forensics scenarios, and optimize the execution speed of the color equalization based forensic tool.