Deep Learning based Model for Deepfake Image Detection: An Analytical Approach
Neha Neha, Bhavna Arora · 2023
Recent years have witnessed great improvement in deepfake technology, spurred by advancements in deep learning models and improved processing capacity. Generative models at the cutting edge of technology have made it possible to produce convincingly realistic synthetic videos, images, and even audio recordings. Deeply fabricated media has the power to harm not just individuals but also our society, institutions, countries, religions, and others. These fake images may circulate online in low quality and contain many forms of distortion that could impair the effectiveness of detection methods. Our study employed a dataset of 140K Real and Fake images to analyse how image distortions affect the detection model using one of the deep learning models, DenseNet. This is accomplished by adding blur and noise to the original dataset and feeding it to the trained neural network to discriminate real and deepfake images. The results demonstrate that the model’s accuracy decreased with the low-quality dataset.