This project implements an automatic image colorization system that converts grayscale landscape images into color images using a deep learning model.
A convolutional encoder-decoder architecture is trained using paired grayscale and color landscape images to learn the mapping between grayscale inputs and their corresponding color images.
The project demonstrates practical applications of:
- Deep Learning
- Computer Vision
- Image Processing
- Convolutional Neural Networks
- TensorFlow and Keras
The main objective of this project is to develop a deep learning model that can automatically predict color information from grayscale landscape images.
The dataset contains paired grayscale and color landscape images.
- Total paired images: 2,171
- Training images: 1,736
- Test images: 435
- Image size: 160 × 160 × 3
- Train/Test split: 80/20
The images were resized and normalized to the range 0–1 before training.
The project uses a convolutional encoder-decoder architecture.
The encoder progressively downsamples the input image using convolutional layers with stride 2.
The decoder reconstructs the image using transposed convolution layers.
Skip connections are used between encoder and decoder layers to help preserve spatial information during reconstruction.
- Python
- TensorFlow
- Keras
- NumPy
- OpenCV
- Matplotlib
- Google Colab
The model was trained using:
- Optimizer: Adam
- Learning Rate: 0.001
- Loss Function: Mean Absolute Error (MAE)
- Epochs: 15
- Batch Size: 15
The model was evaluated on 435 unseen test images.
| Metric | Result |
|---|---|
| Test MAE | 0.04947 |
| Best Validation Loss | 0.0423 |
| Best Validation Epoch | Epoch 14 |
The model successfully generated colorized images from grayscale landscape images.
Visual comparison was performed using:
Grayscale Input → Predicted Color → Actual Color
Sample colorization results are available in the images/ directory.
The results compare the grayscale input, predicted color image, and original color image.
Automatic-Image-Colorization-Using-Deep-Learning/
│
├── image_colorization.ipynb
├── README.md
├── requirements.txt
├── LICENSE
├── .gitignore
│
└── images/
└── project result images