MNIST Digit Classification with CNN

1. Dataset Description

We used the MNIST dataset, which contains 70,000 grayscale images of handwritten digits (0-9), each of size 28x28 pixels.

Class	Number of Samples
0–9	~6000-7000 images per class

---

2. Applied Filter: Thresholding

We applied a Thresholding filter at value 100 to convert grayscale images to binary (black/white).

Purpose:

• Highlight digit structure
• Simplify image features for easier classification by the CNN model.

---

3. CNN Architecture

Layer	Type	Parameters
1	Conv2D (32 filters, 3x3)	Input: (28, 28, 1)
2	MaxPooling2D (2x2)
3	Conv2D (64 filters, 3x3)
4	MaxPooling2D (2x2)
5	Flatten
6	Dense (128 neurons, ReLU)
7	Dense (10 neurons, Softmax)

• Optimizer: Adam
• Loss Function: Sparse Categorical Cross-Entropy
• Epochs: 10
• Batch Size: 32

---

4. Results

• Training Accuracy: ~99.8%
• Validation Accuracy: ~98.86%

---

5. Visualizations

Model Performance

Training and validation accuracy and loss curves across 10 epochs. The model shows rapid convergence with minimal overfitting, achieving high accuracy (~99.7% training and ~98.9% validation), indicating successful learning

---

Sample Digit Grid

A grid of 25 sample images from MNIST, showing handwritten digits (0–9) with their labels to visualize and verify dataset correctness.

---

Thresholding Example

The original and thresholded versions of a handwritten digit "3". Thresholding filter was applied to enhance the contrast and simplify the image, converting it into a binary format (black and white) for better feature extraction.

---

6. Conclusion

Applying the Thresholding filter successfully simplified the input images while preserving essential classification features.
As a result, the CNN model achieved high accuracy, demonstrating the effectiveness of preprocessing (thresholding) in digit classification tasks.

7. Project Files

• img.py: Load and display images.
• minst.py: Example for handling and classifying digit datasets.
• save_images.py: Save processed images to disk.
• python.image1.pdf: Sample exported PDF for results.

8. How to Run

1. Ensure Python is installed.
2. Install the required libraries: pip install matplotlib numpy pillow
3. Run any of the scripts as needed: python img.py

🧑‍💻 Author

• Rodina Ahmed → GitHub Profile
• Farida Ayman → GitHub Profile
• Nada Attia → GitHub Profile