# pid-pipe-removal

**Repository Path**: dlml2/pid-pipe-removal

## Basic Information

- **Project Name**: pid-pipe-removal
- **Description**: No description available
- **Primary Language**: Unknown
- **License**: Not specified
- **Default Branch**: main
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2026-09-09
- **Last Updated**: 2026-09-09

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# 🛠️ P&ID Pipe Removal Project (Final Version)

Automatically removes connecting pipes from engineering P&ID diagrams using OpenCV, while preserving all critical symbols (valves, sensors, labels).

---

## 📅 Project Overview
This project was completed as part of an AI/ML software engineering assignment. The objective was to detect and remove connecting pipes from a P&ID (Piping and Instrumentation Diagram) while keeping all symbols intact. The final solution uses image processing techniques such as grayscale differencing (`absdiff`), morphological operations, and inpainting.

Although the initial attempt did not yield accurate results within the deadline, I later reworked the solution using Google Colab and achieved precise, ground-truth-matching output through iterative debugging and testing.

---

## 📊 Results
| Original Diagram | Processed Output |
|------------------|------------------|
| ![](data/input.jpg) | ![](data/final_output.jpg) |

Pipe connections are removed, and all relevant symbols and texts are preserved with high accuracy after extensive corrections.

---

## 🔄 How It Works
### Step-by-step Pipeline:
1. Load the original input image and the expected pipe-free version (used to guide mask generation).
2. Convert both images to grayscale using OpenCV.
3. Generate a pipe mask by applying `cv2.absdiff()`, followed by thresholding and dilation.
4. Apply OpenCV's `inpaint()` function to remove the pipes while preserving background and symbol details.
5. Save the cleaned output and pipe mask for verification.

This method was finalized after comparing alternatives like contour detection, Hough Transform, and a U-Net model, but OpenCV inpainting with grayscale differencing provided the best accuracy.

---

## 🔧 Technologies Used
| Library | Purpose |
|--------|---------|
| OpenCV | Image preprocessing, grayscale diff, inpainting |
| NumPy  | Image data manipulation |
| Python | End-to-end implementation |
| Google Colab | Final development environment |

---

## 📁 Folder Structure
```
pipe_removal_pid_ai/
├── data/
│   ├── input.jpg                  # Original P&ID diagram
│   ├── nearest_expected_output.jpg # Clean reference for mask generation
│   └── final_output.jpg           # Output image with pipes removed
├── results/
│   └── mask_preview.png           # Binary pipe mask
├── generate_pipe_free_output.py   # Main script
├── README.md                      # Project documentation
```

---

## 🚀 How to Run
### Requirements:
```bash
pip install opencv-python numpy
```

### Run the Script:
```bash
python generate_pipe_free_output.py
```

This will:
- Generate `results/mask_preview.png`
- Generate `data/final_output.jpg`

---

## 🧑‍💻 Author
**Waseem Ibn Yousef CM**  
Calicut, Kerala, India  
📧 cmwaseemyousef@gmail.com  
📞 +91-9746460363  
[LinkedIn](https://linkedin.com/in/waseemibnyousefcm)

---

## 🚀 Showcase Ready
GitHub Repo: [https://github.com/cmwaseemyousef/pid-pipe-removal](https://github.com/cmwaseemyousef/pid-pipe-removal)

---

## 🎥 Video Demo (Script Outline)
1. **Intro**: "Hi, this is Waseem. Here's my AI project that removes pipes from P&ID diagrams."
2. **Problem**: "Engineers often need clean diagrams for automation and data extraction."
3. **Solution**: "I compare the original image with an expected clean version, generate a pipe mask using grayscale difference, and apply OpenCV's inpainting to remove the pipes."
4. **Demo**: Show how it runs in Google Colab and display before/after results.
5. **Outro**: "Thanks for watching! Visit my GitHub to explore the code."

---

## 🔗 License
This project is for educational and demonstration purposes only. Feel free to fork and improve.