Forensic Approach For Object Elimination and Frame Replication Detection Using Noise Based Gaussian Classifier

Abstract

In the modern world, people are beliving videos as part of social communication; as camera editing techniques are advanced, video doctoring is a technique for editing and recreating new details in the footage. Identifying these doctored videos poses a problem for the media source, the court of law, and the framework of evidence service. The research on video forensics, and specifically on the automatic recognition of object-based detection of video forgery, is still in its infancy. The approach proposed in this paper uses noise properties, extracted from each frame of the video using Wavelet Transform and nonlinear thresholding such as optimal SURE shrinkage. Gaussian Mixture Density (GMD) uses this as a Gaussian classifier, and the Expectation-Maxima algorithm sets the GMD parameter. Results of the output matrix show that we get excellent precision 99.36 percent recall 99.80 and precision 97.34 percent respectively for object removal and frame duplication detection compared to subsisting methods. The proposed approach effectively detects traces in the forensic video dataset and recognizes these.