Detection of Quality Deep Fake Images and Videos Using Customised Convolutional Neural Networks
Alap Mahar, Pushpneel Verma, Ajit Singh · 2025
Due to advancements in artificial intelligence there are numerous deep fake images and video collections are available on the Internet and social media. The primary aim of the study is to analyse the deep fake images and videos since the quality is constantly improving a novel method was developed for accurate detection of quality deep fakes. The suggested approach uses a customised convolutional neural network (CNN) technique that uses facial landmark detection to extract structured data from photos and video frames before feeding it into the methods of CNN. The customised CNN model is augment-based CNN for generation of fake images and fake data. Involves about 260 films from the data set in which 202 images are made up while others were real images. About 300 videos are used in which 250 videos are fake and 50 were real the proposed model achieved the accuracy of 95.58% and 0.97 AUC score that outperforms the existing models like MLP-CNN and CNN. Additionally, the method succeeds with greater accuracy then the conventional models like DST -Net, VGG 16 Efficient Net. This research study's primary goal is to create a new CNN learning method for identifying high-quality deep fake photos and videos.