Research On Video Forgery Passive Detection Based On Spatial-Temporal Feature

With the development of computer science and Internet, there comes a variety of computer crimes. The dramatic increase in the amount of video data, fast spread of video, and popularity of video editing software make the tampering of videos easier, which bring more and more problems related to personal privacy, law enforcement investigations and social stability. Hence, video tamper detection has become an important and urgent issue in the field of multi-media content analysis.There are two ways of video forgery detection: active detection method and passive detection method. Active detection approach uses digital watermark or digital signature to detect video forgery. This method is characterized by the need of preprocessing videos before transmission, such as embedding digital watermark or digital signature information. Therefore, the active detection method fails when a given video is not embedded with a digital watermark or signature in advance. In recent years, people mainly focus on video forgery passive detection which does not require watermark and digital signature. The basic principle of passive video tamper detection is to use the inherent statistical characteristics of the video, which are also known as the finger print of videos, to detect whether the video is tampered or not. It is an effective supplement to active detection approach and has aroused more and more attention around the world.This paper gives a comprehensive study of the current passive video forgery detection research both at home and abroad, and summarizes the basic and common video forgery detection model. On the basis, this paper mainly focuses on the following two aspects in depth: On the one hand, in order to detect common video forgeries (frame duplication, frame region copy-move and frame deletion), a new method of exposing video forgeries based on SURF (Speeded Up Robust Features) is proposed.Besides, this paper studies the passive video copy detection technology, which is also called CBCD (Content Based Copy Detection). Detecting video copies is important for copyright control, business intelligence and advertisement tracking, law enforcement investigations, etc. This paper mainly deals with low quality coding, temporal or spatial domain video editing , such as covering objects, inserting logos, adding subtitles, zoom and so on. A content-based video copy detection algorithm that detects illegally copied videos is proposed. Firstly, the spatial-temporal feature is extracted to describe the video. Then comparing the video with the preregistered features stored in the database of videos using k-dimension tree (k-d tree) indexing structure, the proposed approach distinguishes whether the query video is illegally copied or not. At last, a large number of experiments based on the TRECVID CBCD task video database are completed, and the experimental results demonstrate that the proposed method has a good precision and efficiency in detecting the following video copy: simple global transformation, Gamma adjustment, subtitle insert, logo insert and frame zoom-in and zoom-out.