Cannotation MeasureUp to Detect Deepfake by Face Recognition via Long Short-Term Memory Networks Algorithm

Abstract

In this computerized world, any individual can create deep fake videos, images, and audios, and it’s becoming a challenge to differentiate between the real and the fake ones. Such fake videos and images of a person that spread like wildfire on social media ultimately destroys the person's life by leaving a deep mark that is impossible to remove. Hence, it’s important to discern the fake and stop the circulation of these types of videos and images. Different technologies with different algorithms have been used to identify fake images. In this research, to distinguish fake and real images, long short-term memory network methods in deep learning technology are used to recognize faces from real-time dataset for better result because of its hidden layers and some inbuilt functions of this algorithm. Haar Cascade is used for detecting the images from real time; even though it’s an oldest one, it still has its own place in detecting face by providing more accuracy than others. Backpropagation is then used to feed both forward and backward, so that the values of bias can be reduced according to the predicted output. Sequence transformer makes the images get into cell state with continuous step-up time, so that the image will not collide while capturing. By using all these conditions, the output will be predicted with actual value which gives the desired outcome. Detecting the swapped and manipulated images from the real one is very difficult even though different algorithms are used. Therefore, long short-term memory networks are used in this research, because it's easy to identify fake images due to the continuous networks in the layer.