AN OPTIMIZED CONVOLUTION NEURAL NETWORK BASED INTER-FRAME FORGERY DETECTION MODEL - A MULTI-FEATURE EXTRACTION FRAMEWORK

Surveillance systems are becoming pervasive throughout our daily lives, and surveillance recordings are being used as the essential evidence in criminal investigations. The authenticity of surveillance videos is tough to confirm. One of the most popular methods of video tampering is inter-frame forgery. Using an optimised deep learning methodology, a novel inter-frame forgery detection and localization model is introduced in this research work. Pre-processing, feature extraction, and forgery detection will be the three main phases of the presented design forgery detection model. In the detection model, the original video frames will be pre-processed to improve the image quality. The pre-processing phase includes the frame extraction from video, grey conversion and removal of movement frames as well. Following that, features such as SURF, PCA-HOG features, MBFDF, correlation of adjacent frames, PRG, and OFG based features is extracted. These extracted features will be subjected for forgery detection using Optimised CNN with fine-tuned weights by the hybrid approach. The suggested hybrid paradigm Mayfly Optimization espoused Black Widow Optimization (MO-BWO) is a mathematical fusion of both the Black Widow Optimization (BWO) and Mayfly Optimization Algorithms (MA). In case if the video is detected to be prone to tampers, then the corresponding location gets trapped in the localization phase. Moreover, the detection phase will portray the information regarding the type of tamper like duplication, insertion and deletion of frames. Here, the exact tamper localization is accomplished based on the PRG and OFG. Finally, the supremacy of the MO-BWO+CNN is validated over other conventional models.