Video Frame Deletion Detection using Correlation Coefficients

Abstract

In this paper, we propose feature-based machine learning models for detecting frame deletion tampering in videos. The work investigates inconsistency in correlations between adjacent frames that occurs when frames are dropped from a continuous sequence. As a result, the correlation pattern of the original and counterfeit videos differs slightly. Three machine learning models namely Support Vector Machine (SVM), Multilayer Perceptron (MLP) and Convolution Neural Network (CNN) have been implemented to predict the authenticity of video shots. Experiments have been conducted on a large dataset of 600 videos each of 25-frame deletion and 100-frame deletion. The results show that the CNN model can classify between authentic and forged sequences more accurately than SVM and MLP with the highest accuracy of 97% for 100-frame deletion.