Forgery Detection in Ballistic Motion Videos Using Motion Estimation and Modelling

Abstract

AbstractIn this paper, we propose a method for detection of forgery in ballistic motion videos using motion estimation and modelling. Motion between consecutive frames is estimated using block matching algorithm without interpolation and is represented by a motion vector. State matrices are formed from the motion vector and a Markov process model is applied to it to get the transition probability matrix. By analysing the probability values in this matrix, the transition from one frame to the next is evaluated. For authentic videos, the transition probability matrices for all pair of subsequent frames show uniform characteristics whereas for fake videos, we can see difference in these characteristics. Thus transition probability matrix is used here as a feature vector for classifying a ballistic motion video as an authentic one or fake. The method is evaluated using various original and fake ballistic motion videos and yields good results in both static and moving camera videos.KeywordsBallistic MotionVideo Forgery DetectionMotion EstimationMarkov Modelling