Dataset Preprocessing & Mask Generation

Overview

This project prepares the concrete crack dataset for machine learning and deep learning applications.

The preprocessing pipeline includes:

• Image resizing (256 × 256)
• Pixel normalization
• Grayscale conversion
• Data augmentation
  • Horizontal Flip
  • Vertical Flip
  • Rotation (±10–15°)
  • Zoom (5–10%)
• Binary mask generation using OpenCV

---

Features

Dataset Preprocessing

• Resize images to 256 × 256
• Normalize pixel values
• Convert images to grayscale
• Organize dataset into train/test folders

Data Augmentation

• Horizontal Flip
• Vertical Flip
• Rotation
• Zoom

Mask Generation

• Gaussian Blur
• Canny Edge Detection
• Morphological Dilation
• Binary Thresholding

---

Project Structure

SDNET2018-Preprocessing/
│
├── dataset_preprocessing.py
├── generate_masks.py
├── README.md
├── requirements.txt
├── .gitignore
└── output_examples/

---

Installation

Install the required packages:

pip install -r requirements.txt

---

Run Dataset Preprocessing

python dataset_preprocessing.py

---

Generate Masks

python generate_masks.py

---

Output Examples

Example outputs generated by the preprocessing pipeline are available in the output_examples/ folder.

Original Image

Preprocessed Image

Horizontal Flip

Rotated Image

Zoomed Image

Generated Mask

---

Technologies Used

• Python
• OpenCV
• NumPy

---

Dataset

SDNET2018 – A Concrete Crack Image Dataset for Machine Learning Applications.