Deep learning model for deep fake face recognition and detection

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Deep learning, particularly generative adversarial networks (GANs), enables sophisticated image manipulation, creating deep fakes indistinguishable to humans.
• The proliferation of deep fakes poses significant threats to public trust and security.
• Existing deep fake detection methods often suffer from inaccuracies and high computational costs.

Purpose of the Study:

• To develop an accurate and efficient deep learning-based method for detecting deep fake face images.
• To address the limitations of current deep fake detection techniques, such as inaccuracy and high processing time.

Main Methods:

• Implementation of a deep learning technique combining Fisherface with Local Binary Pattern Histogram (FF-LBPH) for face recognition and dimensionality reduction.
• Application of Deep Belief Networks (DBN) with Restricted Boltzmann Machines (RBM) as a classifier for deep fake detection.
• Utilizing public datasets such as FFHQ, 100K-Faces DFFD, and CASIA-WebFace for model training and evaluation.

Main Results:

• The proposed FF-LBPH method, integrated with DBN and RBM, demonstrates effective deep fake face image detection.
• The approach aims to overcome the inaccuracies and time-consumption issues prevalent in existing detection methods.

Conclusions:

• The developed deep learning approach shows promise in enhancing the reliability of deep fake detection.
• Further research in this area is crucial to combat the evolving threat of manipulated digital media.