This project aims to detect fake or manipulated brand logos using pattern recognition techniques implemented through a Convolutional Neural Network (CNN).
The model is designed to classify input logos as Real or Fake by learning distinctive visual patterns that separate authentic logos from counterfeit ones.
The dataset is sourced from Kaggle, containing a wide range of real and fake brand logos such as Google, YouTube, Fila, LG, and Levi’s.
Dataset Structure:
/train/real → Real brand logos
/train/fake → Fake or generated logos
Each image is resized and normalized before being fed into the CNN model for training and testing.
A Convolutional Neural Network (CNN) is trained to extract and recognize spatial features from logo images.
The network automatically learns visual differences between authentic and fake logos through deep pattern recognition.
Key Layers:
Conv2D– for feature extractionMaxPooling2D– for dimensionality reductionFlatten– to convert feature maps into a vectorDense– for classificationDropout– to prevent overfitting
Activation Functions:
- ReLU for hidden layers
- Sigmoid or Softmax for output layer
- 🐍 Python
- 🧩 TensorFlow / Keras
- 💻 Google Colab
- 📊 NumPy, Matplotlib, Scikit-learn
- 🗂 Kaggle Dataset
| Metric | Accuracy |
|---|---|
| Training Accuracy | ~95% |
| Validation Accuracy | ~90% |
The model achieves strong classification accuracy, effectively distinguishing between real and fake logos based on visual features.
In this including .ipynb file