Paper Details

Name: Neha Saini
Country: India

Year: 2025
Volume: Volume-12, Issue-2 (July-December)
Page Number: 544-550
DOI: https://doi.org/10.5281/zenodo.21484374

ABSTRACT
The creation of digital media has been transformed by the quick development of artificial intelligence (AI) and deep learning, which has made it possible to produce extremely lifelike synthetic images and movies known as "deepfakes." Deepfake technology poses serious problems with disinformation, identity theft, cybercrime, political manipulation, and digital fraud, despite its many positive uses in entertainment, education, healthcare, and the production of digital material. Consequently, the creation of precise and dependable deepfake detection systems has grown in importance as a field of study in multimedia forensics and computer vision.This research proposes a deep learning-based system for recognizing real and false photos by integrating effective image preprocessing approaches with convolutional neural networks (CNNs) for spatial feature extraction. The suggested method uses Long Short-Term Memory (LSTM) networks to capture temporal relationships between successive video frames for video-based deepfake identification. Data gathering from publicly accessible deepfake repositories, preprocessing with Error Level Analysis (ELA), image normalization, data augmentation, feature extraction, model training, and performance assessment are all part of the methodology. The robustness of the suggested model is assessed using a number of benchmark datasets, such as FaceForensics++, Celeb-DF, DFDC, CIFAKE, and CASIA.Standard assessment criteria like accuracy, precision, recall, F1-score, and confusion matrix analysis are used to evaluate the performance of the suggested system. The findings of the experiments show that deep learning models can detect tiny visual abnormalities and temporal inconsistencies introduced during the creation of deepfakes with excellent detection accuracy across various modification strategies. The report highlights future research paths like explainable AI, lightweight detection models, multimodal learning, and real-time deployment while also discussing present problems like adversarial attacks, limited dataset diversity, and the quick growth of generative models. Overall, by offering an efficient deepfake detection framework, this study helps to enhance the dependability, security, and authenticity of digital material.

Keywords: Deepfake, Image Forgery Detection, Artificial Intelligence, Deep Learning, CNN, GAN, Deep Learning Techniques, LSTM