FORENSIC LENS: DECODING THE

TRUTH THROUGH COMPREHENSIVE

IMAGE ANALYSIS
Fake Photo Analysis and Forensic Examination

PRESENTED BY UNDER THE GUIDANCE UNDER THE SUPERVISION

BOLLAM SUNIL DR V SAMESWARI Mr. G. Vimal Kumar,

MSCS22R002 GUEST FACULTY(ON CONTRACT) CTO
DEPARTMENT OF COMPUTER SCIENCE DEPT OF COMPUTER SCIENCE Cyber Privilege Pvt Ltd
(CYBER SECURITY) (CYBER SECURITY) Hyderabad
 ABSTRACT
Cyber forensics demands a multifaceted skill set to effectively
navigate the complexities of digital investigations. Beyond traditional
forensic techniques, proficiency in various roles such as Techniques
and Procedures (TTPs), Social Media Intelligence (SOCMINT),
Geographic Intelligence (GEOINT), and business due diligence is
indispensable. Moreover, conducting Geospatial Analysis, Corporate
Intelligence, Geopolitical monitoring, Physical Security Assessment,
and Terrorism Analysis are essential components.
 PROBLEM
Problem and Motivation:
• The rise of deepfakes and sophisticated image manipulation techniques necessitates robust
detection methods.
• Manual detection is becoming ineffective due to the evolving nature of manipulation
techniques.
Machine Learning for Image Detection:
• Supervised learning algorithms are trained on labeled datasets (real vs. fake images).
• These algorithms learn to extract features from images that differentiate real from manipulated
ones (lighting inconsistencies, shadows, textures).
• Trained models can then classify new images based on the learned features.
Artificial Neural Networks (ANNs) for Feature Extraction:
• Convolutional Neural Networks (CNNs) are a powerful type of ANN suited for image recognition.
• CNNs excel at extracting features through convolutional layers that use filters to learn patterns
directly from image pixels.
• By training a CNN on real and fake images, the network learns to identify inconsistencies
introduced by manipulation.
 WHY DO WE NEED THIS

 Issue:

Online platforms are increasingly plagued by fraudulent activities involving

fake images.
 Challenges:
Fraudsters use manipulated visuals to deceive users for financial gain or
misinformation.
 Impact:
Trust and credibility on social media are undermined, affecting user
experience and platform reputation.
 Visual:
Display examples of fake images (e.g., deepfakes, digitally altered photos) to
illustrate the severity of the problem.
 IMAGE OSINT FORENSIC TOOLS
 Image Search Tools:
• Google Images: Excellent for finding similar images (https://images.google.com/)
• TinEye: Powerful for identifying original source of an image (
https://blog.tineye.com/category/image-search/)
• Yandex Images: Another option with good international coverage (
https://yandex.com/images/)
 Face Detection and Recognition:
• OpenCV: Open-source library with face detection capabilities (https://opencv.org/)
• Microsoft Azure Cognitive Services: Cloud-based facial recognition with high accuracy (
https://learn.microsoft.com/en-us/azure/ai-services/computer-vision/identity-api-reference
)
• Amazon Rekognition: Similar cloud-based service from Amazon Web Services (
https://aws.amazon.com/rekognition/)
 Image Analysis and Restoration:
• FotoForensics (FREE): Basic image analysis for potential manipulation detection (
https://fotoforensics.com/)
• Adobe Photoshop (PAID): Industry-standard for image editing and restoration (
https://www.adobe.com/products/photoshop.html)
• InPix Forensics (PAID): Advanced software for image forensics analysis (
https://play.google.com/store/apps/details?id=com.imagin.peeku&hl=en_IN)
 Metadata Editing:
• ExifTool (FREE): Open-source tool for viewing and editing image metadata (
https://exiftool.org/)
• MediaInfo (FREE): Versatile tool for displaying technical information from various
media files (https://mediaarea.net/en/MediaInfo/Download)
 QR and Barcode Tools:
• ZXing (FREE): Open-source library for reading and writing various barcode formats (
https://zxing.org/w/decode.jspx)
• Bar Code Scanner Apps: Many free and paid mobile apps can scan QR and
barcodes.
 CHALLENGES
Challenges with CSI and OSNIT Tools:
o Limited to Known Signatures: CSI tools rely on predefined signatures and may struggle with
novel or evolving forgery techniques.
o Manual Feature Engineering: OSNIT tools often require manual feature selection and
extraction, limiting adaptability to diverse forgery patterns.

Rise of deepfakes and manipulated images.

o Negative impacts on trust and information integrity.
o Difficulty in manual detection.
o Image: A split image. One side shows a real image of a person smiling, and the other side
shows a deepfake of the same person with a different facial expression.
PROPOSED WORK
 Fake Photos Detection:
• Fake Image Detection Websites: Several online tools claim to detect
deepfakes, but accuracy varies greatly. Be cautious and use them for
initial screening only.
• Deep learning libraries like TensorFlow or PyTorch: These can be used to
build custom deepfake detection models.

 Advantages of Deep Learning:

• Automated Feature Learning: Deep learning models autonomously learn
discriminative features, reducing dependency on predefined signatures.
• Generalization: Deep learning models excel in generalizing across diverse
forgery types, adapting to new and evolving techniques.
• End-to-End Learning: Ability to learn complex representations directly
from raw data, enhancing flexibility and adaptability.
ARCHITECTURE
 METHODOLOGY

Deep Learning-based Forgery Detection CSI Image Forensic Tools

 Complexity Handling  Feature Dependency
Learns intricate, non-linear patterns. Relies on specific handcrafted features.
Effective for modern, sophisticated forgeries. Limited by feature extraction methods.
 Feature Learning  Adaptability
Automatically extracts relevant features Requires manual adjustment and tuning.
from raw data. Less effective against evolving forgery techniques.
Generalizes well across diverse forgery types.  Performance
 End-to-End Learning May struggle with complex, non-linear forgery
patterns.
Directly maps input images to forgery
classifications. Lower accuracy compared to deep learning
models.
Minimizes manual feature engineering.
 PROGRESS OF WORK
Data Collection
• Gathered a diverse dataset comprising synthetic and real images.
• Ensured sufficient diversity to capture various types of manipulations.
Preprocessing
• Conducted data preprocessing to enhance model performance.
• Applied techniques like normalization, augmentation, and resizing.
Model Development
• Designed a deep learning architecture using Convolutional Neural Networks
(CNNs).
• Experimented with different CNN architectures (e.g., ResNet, VGG) for optimal
performance.
Training and Validation
• Trained the model on the prepared dataset using a designated training-validation
split.
• Monitored training metrics such as loss and accuracy to assess model convergence.
FAKE PHOTO RECOGNITION USING ERROR LEVEL ANALYSIS & CNN
 CONCLUSION
In conclusion, the combination of Error Level Analysis (ELA) and Convolutional Neural
Networks (CNNs) is a promising approach for image tampering detection.

This approach can be used to identify a wide range of image tampering techniques,
including photoshopping, splicing, and copy-paste forgery. However, further research is
needed to improve the accuracy and robustness of this approach.

For example, new techniques could be developed to address the challenges of detecting
complex forgeries