DeepSight: Enhancing Deepfake Image Detection and Classification through Ensemble and Deep Learning Techniques

• Published in: 2024 5th International Conference on Image Processing and Capsule Networks (ICIPCN) pp. 28 - 35
• Main Authors: Manju, T, Kalarani, S
• Format: Conference Proceeding
• Language: English
• Published: IEEE 03.07.2024
• Subjects: Accuracy; Deep learning; Deepfake image; Deepfakes; Ensemble learning; Feature extraction; Information age; Machine learning; Vectors; Visualization
• DOI: 10.1109/ICIPCN63822.2024.00014

in today's digital age, deepfake images poses a significant threat to multimedia content authenticity and integrity. Detecting and classifying deepfake images with high accuracy is crucial to addressing this growing challenge. Deepfake images, generated using advanced machine learning techniques, have become a significant concern due to their potential to deceive individuals and manipulate information. This paper presents DeepSight, an innovative approach that combines ensemble learning techniques with deep learning models to enhance deepfake image detection and classification. DeepSight leverages the diversity of multiple classifiers trained on distinct feature representations extracted from the input image data. The framework begins with an initial assessment to determine whether the image has been manipulated. Subsequently, the image goes through deep feature extraction using Convolutional Neural Networks (CNNs). The resulting feature vectors are then classified using Random Forest, KNearest Neighbors, and XGBoost, with hyper-parameter optimization. Experimental evaluations on benchmark datasets demonstrate DeepSight's superior performance compared to state-of-the-art methods, achieving higher accuracy and robustness across various deepfake generation techniques and quality levels. Notably, DeepSight achieves the highest accuracy of 97.5% when utilizing DenseNet and XGBoost. Overall, DeepSight represents a significant advancement in deepfake image detection and classification. It provides a reliable solution to address deceptive visual content in digital environments.