A New Approach to Detect Deepfake Video using Multi-Input Convolutional Neural Network

In this modern age, Deepfake videos are spreading around, having a severe impact on the social and personal lives of the general public. Efficient techniques, operating better than others, is hence in the requirement. In this research, combination of many objects like left eye, right eye and mouth shapes from image frames of videos are taken into consideration, thus achieving better accuracy than other detection techniques. In this proposal, deep learning techniques, specially the multi-input Convolution Neural Network, are executed. The proposal firstly detects the face, followed by specific objects such as left eye, right eye and mouth. Then multi-input CNN is applied that classifies the data based on eyes and mouth. After applying multi-input CNN, the accuracy increases impressively. Novel techniques are used to extract specific features like eyes, mouth. Memory storage was a concern for the proposal. Despite this, comparison with other relevant research works proved better accuracy and data analysis.