Deepfake Detection using Deep Learning with InceptionV3
K. Cheritha, S. Akhil, V. Bhanu Prakash and A. Akhil Reddy
Dept Of Computer Science and Engineering, Malla Reddy University, Hyderabad
Mrs. Aurchana. P, Assistant Professor.
Dept Of Computer Science and Engineering, Malla Reddy University, Hyderabad
Abstract - Deepfake technology has rapidly evolved, making it increasingly difficult to distinguish between real and manipulated videos. This poses serious risks, including misinformation, identity theft, and digital forgery. To address this challenge, we propose a deep learning-based deepfake detection model that leverages InceptionResNetV2, a hybrid architecture combining the strengths of Inception networks and Residual networks (ResNet). Our approach efficiently extracts key facial features from video frames and classifies them as real or fake. The detection pipeline includes video preprocessing, frame extraction, feature extraction using InceptionResNetV2, and classification through a deep learning model. Our approach utilizes RGB colour space for feature extraction, ensuring that fine-grained visual artifacts present in manipulated videos are effectively captured. The detection pipeline involves video preprocessing, frame extraction, RGB-based feature extraction using InceptionResNetV2, and classification using a deep learning model. We train and evaluate our model using benchmark deepfake datasets, including Face Forensics++, Deep Fake Detection Challenge Dataset, and DeeperForensics-1.0, which contain diverse real and fake video samples. The model predicts whether a video is real or fake while providing a confidence score for better interpretability Experimental results demonstrate that our InceptionResNetV2-based model achieves high performance in deepfake detection. Our model achieves an accuracy of 81.3%, a precision of 82.84%, a recall of 84.00%, and an F1-score of 83.42.1%, indicating its effectiveness in distinguishing between real and fake videos. Future enhancements include real-time detection capabilities, adversarial training for improved robustness, and explainable AI techniques to provide greater transparency in deepfake detection. This research contributes to ensuring the authenticity of digital content and strengthening defences against deepfake-based cyber threats
Keywords – DFDC Deepfake Detection Challenge, CNN-Convolutional Neural Networks
