CLH: Approach for Detecting Deep Fake Videos

Deep Fakes are the media that takes the person’s image in an existing photograph, audio recording, or video and replaces them with another person’s likeness by making use of synthetic intelligence and device mastering. In this era, everybody can get easy access to software packages and tools to create deep fake videos. Existing techniques are constructed with the usage of the lip synchronization, mouth features artifacts and are commonly designed for detection of single frames. The proposed model, CLH (CNN+LSTM hybrid model) considers various parameters such as eye blinking, blurriness, skin tone, skin color, changes in lighting, lip syncing, and position to detect the fake videos. The CLH model employs “Convolutional Neural Networks (CNN)” and “Long Short-Term Memory (LSTM)” for detecting a deep fake video. The original videos and deep fake (high quality + low quality) videos were used in training the model. Datasets such as Celeb-DF, face forensics ++, Deep fake TIMIT, and fake videos developed by Facebook were used to train and evaluate the model, so that an efficient model is constructed. The proposed CLH model achieved a high accuracy of more than 90% and a low false positive rate of less than 5%. The CLH model is also compared with other models on the market and analyzed to understand the significance of the work.