Improved Copy-Move Forensics Techniques for Digital Video

Abstract

ABSTRACT There are a lot of editing tools, software and devices like mobile phone, digital cameras, digital camcorders, etc., that are easily used to manipulate the contents of authentic digital videos without leaving behind any footprints. During manipulation, any frame or region of digital video is altered with Copy-Move (CM) forgery, which results in forged digital videos with Copy-Move Frame Duplication (CMFD), Copy-Move Region Duplication (CMRD), and Chroma Key Foreground (CKF) forgeries. Due to these forgeries, it becomes more difficult to detect the authenticity of digital videos through the eyes. Therefore, it is essential to ensure digital videos' reality in this real-world scenario. There are several existing techniques for detecting these forgeries in digital videos in the literature. However, these existing techniques have suffered from several limitations for detecting the CMFD, CMRD and CKF forgeries in digital videos. In this thesis, work has been done to improve the detection of CMFD, CMRD and CKF forgeries in digital videos. In the presented work of CMFD forgery detection, the proposed approach has detected duplicated frame sequence at long continuous locations as well as at many different locations with different lengths of frame sequence in the digital videos using Correlation Coefficients (CC) and also detected the frame sequences which are duplicated from other digital videos with Coefficient of Variation (CV). Thus, it is observed that the presented work is more effective for detecting frame duplications in large and small frame sequences with higher detection accuracy (DA) in digital videos than the existing techniques. The limitation of detecting the less number of duplicated frames has been removed in multiple CMFD forgery detection approach. This presented work is based on Equal Central Block Variance (ECBV), which has efficiently detected multiple frame duplications in the digital videos such as single frame duplication (SFD) in the entire digital video, repetition of a frame (RF) in the form of sequence, shuffled frame sequence (SFS) and disorder frame sequence (DFS). By comparing the results, it has been found that early and effective detection with higher DA and minimum execution time achieved in different digital video cases by multiple CMFD forgery detection approach. In the presented work for detecting CMRD forgery, the CMRD forgery detection approach has detected rectangular and square shape regular region duplications and irregular region duplications with many irregularities within the same frames and from the other frames of the digital videos. This presented work is based on CC and CV to detect regular and irregular region duplications in digital videos. From its comparison with existing techniques, it has been noticed that the proposed approach provides more effective performance with higher DA in digital videos taken from the SULFA dataset and downloaded from the internet. The drawback of detecting small duplicated regions in the digital videos has been eliminated in the proposed multiple CMRD forgery detection approach based on Histogram Equalization (HE) and block filtering. This presented work has effectively detected single and multiple CMRD forgeries with different region sizes such as 3× 3, 4× 4, 8×8, 16×16, 24× 24, and 32× 32 in the digital videos. It is observed by comparing this work with the existing techniques that it provides better results on detecting single and multiple CMRD forgeries with different region sizes in the digital videos than others. There are very few existing techniques for detecting CKF forgery in the literature. In this thesis, the CKF forgery detection approach has been seen this forgery based on Frame Edge Identification (FEI) in the digital videos. The proposed method has identified and detected CKF forgery within the edge frames. This approach has also isolated this forgery from the authentic part of the edge frame with localization and tracking in each frame of the forged digital video. This presented work has performed more efficiently on the digital videos with different cases taken from the SULFA dataset. This work has provided adequate robustness against various attacks. It has also implemented digital videos, which are downloaded from the internet. The experimental results indicate higher DA with lower execution time and better robustness than existing techniques.