High Impact Factor : 4.396 icon | Submit Manuscript Online icon |

Detecting Fake and Fraudulent Faces in Images and Videos Using a Convolutional Neural Network and a Django Web Interface

Author(s):

Yeshaswini Bhandari K R , ATME College of Engineering; Vijay Kumar S, ATME College of Engineering; Harshitha H B, ATME College of Engineering

Keywords:

Fake Face Detection, Deepfake, Convolutional Neural Network, TensorFlow, Django, Grad-CAM, Image Forensics

Abstract

Advances in image synthesis now allow highly realistic fake faces to be produced, creating risks of disinformation, identity fraud and loss of trust in online media. This paper presents a prototype system that classifies face images and video as real or fake. A convolutional neural network (CNN) with three convolution–pooling stages and a sigmoid output is trained on a labelled directory of real and fake face images using the Adam optimiser and binary cross-entropy loss. Face detection and a pre-trained face-embedding network with Grad-CAM visualisation support face extraction and interpretation. The trained model is served through a Django web application in which a user uploads a video (MP4, MOV, WMV or AVI) or selects a trending online video and receives a REAL or FAKE verdict. Over eight training epochs the training accuracy rose to roughly 85% while the validation accuracy reached roughly 66%, and the validation loss stopped improving after the fifth epoch, which shows overfitting on the available data. The paper discusses these results honestly, describes the testing approach, and outlines how larger datasets, regularisation and memory-based architectures can improve generalisation to unseen manipulations.

Other Details

Paper ID: IJSRDV14I90065
Published in: Volume : 14, Issue : 9
Publication Date: 01/12/2026
Page(s): 140-143

Article Preview

Download Article