Table content

1. Introduction……………………………………………………………PAGE 2
• Definition of fake image
• Types
• Characters
• Implementation
2. Background &Related Work……………………………………….PAGE 3
3. Categories of Fake Images…………………………………………..PAGE 3
4. Detection Techniques………………………………………………..PAGE 3
• Traditional Approaches
• AI-Based Approaches
5. Proposed method……………………………………………………....PAGE 4
• Steps
• How it Works
• Image manipulation and Re-Detection
6. Implementation……………………………………………………....PAGE6
• Key Tools & Libraries
7. Results & evaluation…………………………………………………PAGE 6
8. Conclusion……………………………………………………………PAGE 10

1|Page
1. Introduction

The rise of fake images—ranging from deepfakes to AI-generated visuals—has introduced new
threats to digital trust, especially in news, social media, and security systems. These images are
often used to spread misinformation, manipulate public opinion, or commit fraud. This project
aims to build a deep learning-based system to detect and classify fake images accurately and
efficiently.

Definition of fake image

A fake image is a manipulated or artificially created visual representation that is intended to

deceive or mislead people into believing it is real.

Types of Fake Images

1. Photo editing: Using software like Adobe Photoshop to manipulate or alter existing images.
2. Deepfakes: Using artificial intelligence (AI) and machine learning (ML) to create realistic
images or videos that are entirely fabricated.
3. AI-generated images: Creating images from scratch using AI algorithms, such as generative
adversarial networks (GANs).
4. Image synthesis: Combining multiple images or elements to create a new image.

Characteristics of Fake Images

1. Inconsistencies Fake images may contain inconsistencies in lighting, shadows, or textures.
2. Artifacts: Fake images may exhibit artifacts, such as unnatural or distorted features.
3. Lack of context: Fake images may be presented without context or with misleading captions.

Implications of Fake Images

1. Misinformation: Fake images can spread false information or propaganda.
2. Identity theft: Fake images can be used to impersonate individuals or create fake identities.
3. Fraud: Fake images can be used to commit fraud or deceive people for financial gain.

2|Page
2. Background & Related Work

Image manipulation techniques have evolved from basic photo editing to sophisticated forgeries
using machine learning.

Traditional methods involved manual detection or forensic tools, but modern fake image
generation—particularly with Generative Adversarial Networks (GANs)—requires AI-based
detection.

3. Categories of Fake Images

• Copy-Move Forgery: A region of the image is copied and pasted elsewhere within the
same image to hide or duplicate elements.
• Image Re-editing: New element are added to original image.
• GAN-Generated Images: Entirely synthetic images created using GANs, often
indistinguishable from real photos to the human eye.
• Deepfakes: Faces or entire videos are manipulated using AI, often replacing one person’s
face with another.

4. Detection Techniques

Traditional Approaches:

- Manual Detection: Visual inspection for inconsistencies in lighting, shadows, and textures.

AI-Based Approaches:

• Convolutional Neural Networks (CNNs): Automatically learn patterns and features

from image data
• Transformer Models (e.g., ViT): Capture long-range dependencies and patterns
• Frequency Domain Analysis: Detects unnatural frequency patterns introduced by GANs
• GAN Fingerprinting: Identifies hidden artifacts or model-specific traces

5. Proposed Method

Face Detection and Verification Using OpenCV

3|Page
In the proposed method, we utilize OpenCV, a widely used open-source computer vision library,
to detect and verify human faces within an image. The goal is to demonstrate how facial regions
can be identified and how image manipulation (e.g., masking or tampering with faces) affects
detection accuracy.

Step 1: Detecting Faces in an Original Image

We begin by uploading an image and detecting the number of visible human faces. This is done
using a pre-trained Haar Cascade classifier, which is effective in identifying frontal human faces.

Here is the Python code used for this process in Google Colab:

# Install OpenCV if not already installed (usually it's pre-installed in

Colab)
!pip install opencv-python-headless

import cv2
import matplotlib.pyplot as plt
from google.colab.patches import cv2_imshow
from google.colab import files
import numpy as np

# Download the pre-trained Haar Cascade classifier

!wget
https://github.com/opencv/opencv/raw/master/data/haarcascades/haarcascade_
frontalface_default.xml

# Load the cascade classifier

face_cascade =
cv2.CascadeClassifier('haarcascade_frontalface_default.xml')

# Function to detect faces in an image

def detect_faces(image_path):
# Read the input image
img = cv2.imread(image_path)

# Convert to grayscale (face detection works on grayscale images)

gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

# Detect faces
faces = face_cascade.detectMultiScale(gray, 1.1, 4)

# Draw rectangles around the faces

for (x, y, w, h) in faces:
cv2.rectangle(img, (x, y), (x+w, y+h), (255, 0, 0), 2)

4|Page
 # Display the output
plt.figure(figsize=(10, 10))
plt.imshow(cv2.cvtColor(img, cv2.COLOR_BGR2RGB))
plt.axis('off')
plt.show()

# Display verification message

if len(faces) > 0:
print("\033[1;32mTHIS IMAGE IS ORIGINAL\033[0m") # Green text
print(f"Number of faces detected: {len(faces)}")
else:
print("\033[1;31mTHIS IMAGE IS FAKE\033[0m") # Red text

# Upload an image file

uploaded = files.upload()
image_path = next(iter(uploaded))

# Detect faces in the uploaded image

detect_faces(image_path)

---
How It Works:

1. Install and import dependencies: This includes opencv-python-headless, matplotlib, and Colab
tools.
2. Load the Haar Cascade classifier: This XML file contains data used to identify human facial
features.
3. Read and process the image:

The uploaded image is read and converted to grayscale, which simplifies the processing for face
detection.

How the uploaded image is displayed

4. Detect and mark faces:

Detected faces are highlighted using blue rectangles.
5. Verification output:

If faces are detected, it prints “THIS IMAGE IS ORIGINAL” in green.

If no faces are found, it prints “THIS IMAGE IS FAKE” in red.

5|Page
---

Image Manipulation and Re-Detection

After the first detection, the next step is to simulate tampering by editing the image and covering
the face regions (e.g., with black boxes or blur). The edited image is then uploaded again, and the
detect_faces() function is re-run.

This helps verify whether the system can still detect faces or recognize the image as potentially
manipulated. A failure to detect faces in a previously verified image indicates the image may
have been altered.

6. Implementation

Key Tools & Libraries

Purpose Preferred Options Why They Matter

Core face- OpenCV-Python (opencv- Fast C++ backend, huge community, well-tested
detection / CV python-headless) Haar & DNN face detectors.

Notebook runtime Google Colab (or Jupyter) Free GPU/CPU, zero setup, easy file upload.

7. Results & Evaluation

Below are some Executed images which are uploaded to detect the faces then re-uploaded after the re-
editing of the original image.

6|Page
IMAGE 1 – Original image

The output
It displays the number of faces it detected, with a blue rectangle on the face it detected
and write out the
THE IMAGE IS ORIGINAL
NUMBER OF FACE DETECTED 1

7|Page
IMAGE 1 – Re-edited image

The output
It displays that no face is detected and write out
THE IMAGE IS FAKE

8|Page
IMAGE 2 – Original image

The output
It displays the number of faces it detected, with a blue rectangle on the face it detected
and write out the
THE IMAGE IS ORIGINAL
NUMBER OF FACE DETECTED 2

9|Page
IMAGE 2 – Re-edited image

The output
It displays that no face is detected and write out
THE IMAGE IS FAKE

8. Conclusion

Conclusion

In conclusion, the detection of fake images is a critical aspect of maintaining digital safety and
trust in today's world. As forgery techniques continue to evolve, it is essential to develop and
improve detection methods to keep pace. This project presents a deep learning-based approach to
detecting fake images, contributing to the ongoing effort to combat digital misinformation.

10 | P a g e
Key Takeaways:

- Fake image detection is a complex task that requires advanced techniques and algorithms.

- Deep learning-based approaches have shown promising results in detecting fake images.

- Continual research and innovation are crucial for maintaining digital safety and trust.

11 | P a g e