Detection of Fake Images using Machine Learni...

Detection of Fake Images using Machine Learning

Author Name : Razik Singh Grewal

ABSTRACT

Images are often manipulated with the intent and purpose to benefit one of the parties. In fact, images are often considered as evidence of a fact or reality, therefore, fake news or any form of publication that uses images that have been manipulated in such a way have the capability and greater potential for misleading. In addition, false news can affect stock prices, which could benefit those who publish news. Another reason is to gain the support or support of other political or political parties. To detect falsification of the image, image data is needed in large quantities, and many models are required to process each pixel in the picture. In addition, efficiency and flexibility in data training is also needed to support its use in everyday life. Big data and deep learning concepts are the perfect solution for this problem. Convolutional Neural Network (CNN) uses two convolutional layers, one layer of MaxPooling, one layer of fully connected, and one layer of softmax output can reach 91.83% accuracy. The use of the Error Level Analysis (ELA) can increase efficiency and reduce the cost of accounting for the training process by reducing the number of epochs required. Therefore, using architecture CNN which utilize ELA, forgery detection image can reach 91.83% and convergence only with 9 epochs.