Detection Of Counterfeit Images Through Deep Learning based Yolov3 Model

Image forgery is the procedure of analysing and detecting any fraudulent or unauthorised alterations made to digital images. The detection in images is difficult for achieving a high accuracy rate and the process is time-consuming. Researchers and forensic experts are developing tools for detecting fake images. The identification of computer-generated picture forgeries is a notable area of study, aiming to provide a comprehensive review of detection tools and techniques. In this work, a novel deep-learning model is proposed for detecting forgery in images. The images are pre-processed using an adaptive Gaussian bilateral (AGB) filter to enhance the quality of the images. The enhanced photos are fed into the Mobile Net for extracting the relevant features in the pictures. The modified Yolov3 is utilized for object detection in images and detecting the real and forgery images. According to the result, the proposed model attains 99.55% for forgery detection in images. The modified Yolov3 yields 99.55% accuracy as opposed to Mask RCNN, RCNN, and SSD's 0.68%, 0.65%, and 1.16% accuracy. CNN, ResNetV2 and DCNN are all outperformed the proposed model in terms of overall accuracy by 0.11%, 1.10%, and 5.55%, respectively.