IMPROVING REALISM IN FACE SWAPPING USING DEEP LEARNING AND K-MEANS CLUSTERING

Abstract

Facial swapping technology is a rapidly growing area of research with a wide range of applications, including entertainment, security, and healthcare. In this project, a deep learning approach was used to achieve highly accurate and realistic face swaps. Specifically, a CNN encoder and decoder network was trained using a large dataset of facial images, and facial clusters were generated using k-means clustering. Computer vision methodologies were also employed to accurately detect and align facial landmarks. The resulting model achieved impressive accuracy of 97% to 99.48% in different epochs, demonstrating its potential for various applications. The system configuration for executing the project included an 11th generation Intel Core i7 processor and 16GB RAM, which provided sufficient computational resources for the task at hand. Overall, this project highlights the power and potential of deep learning techniques for generating highly accurate and realistic facial swaps.

Keywords

• face swapping
• deep learning
• convolutional neural network (cnn)
• computer vision
• accuracy