TraceFake: AI-Based Image Authenticity Checker

Project Domain / Category

Artificial Intelligence (AI) / Image Forensics / Deepfake Detection

Abstract / Introduction

With the rise of powerful AI tools like Google Gemini, DALL·E, and Midjourney, it's become increasingly easy to generate highly realistic images that appear real but are entirely synthetic. This raises serious concerns about visual misinformation, digital manipulation, and trust in media content. The goal of this project is to build an AI-based image authenticity checker that can distinguish between real and AI-generated (fake) images. The system will use a combination of techniques such as CNN-based fake image detection, GAN fingerprint analysis, and EXIF metadata inspection. Students will train or use pretrained models to classify images and optionally add visual forensics like ELA (Error Level Analysis) to detect tampering or manipulation. This project aims to develop a standalone tool that can analyze an image and report whether it is likely to be fake or real, along with confidence scores and forensic cues.

Functional Requirements:

FR1: Upload or input an image file to the system.

        FR2: Classify the image as Real or AI-generated using a CNN or pretrained model.

        FR3: Optionally extract and analyze EXIF metadata (e.g., camera info, editing software).

        FR4: (Optional) Apply Error Level Analysis (ELA) to detect compression artifacts or tampering.

        FR5: Display results including prediction, confidence score, and supporting forensic cues.

        FR6: (Optional for bonus) Allow reverse image search integration (e.g., Bing/TinEye) to trace similar images.

Tools:

        Programming Language: Python

        Libraries & Frameworks: TensorFlow, Keras, OpenCV, exifread, Matplotlib

        Models/Datasets:

        StyleGAN/ProGAN-generated images (Fake)

        CelebA or FFHQ (Real)

        Optional: DFDC or NIST GAN Dataset

Optional Tools:

        Streamlit or Flask (for user interface)

        TinEye/Bing API (for reverse image search)

        ELA toolkit (for tampering detection)

        Development Environment: Google Colab / Jupyter Notebook

Supervisor:

Name: Sonia Salman

Email ID: sonia.salman@vu.edu.pk

MS Teams ID: sonia_salman@outlook.com