# k-means-clustering-defect-inspection

**Repository Path**: lou_wei_yao/k-means-clustering-defect-inspection

## Basic Information

- **Project Name**: k-means-clustering-defect-inspection
- **Description**: No description available
- **Primary Language**: Unknown
- **License**: Not specified
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2020-08-24
- **Last Updated**: 2020-12-19

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# Thermography image segmentation algorithm with de-trend filter and clustering 
## Original paper: 
1. https://www.sciencedirect.com/science/article/pii/S0963869515000961 

2. https://www.spiedigitallibrary.org/conference-proceedings-of-spie/9861/98610S/Thermography-and-k-means-clustering-methods-for-anti-reflective-coating/10.1117/12.2222313.full?SSO=1
## Introduction
### Objective
This work applies thermography (Thermal infrared imaging) on different kinds of materials to identify defects inside of them. The objective is to segment the defect area from the non-defect area automatically without human intervention—a machine vision approach.
### Theoretical background
The algorithm assumes that the defect area would have different thermophysical properties (heat capacity and thermal conductivity) than the non-defect area, thus their surface temperature response became different under thermal excitation. The algorithm identified such difference and segmented areas with a distinctive thermal contrast. 
## Algorithm description
Algorithm is designed as follows:
![Fig 1-1 Algorithm flow](https://github.com/zhouxf53/k-means-clustering-defect-inspection/blob/master/1.png)

1. A de-trend filter was applied to remove background noise due to uneven heating and camera itself. 

2. Edge detection algorithm was applied to highlight the boundary with sharp gradient.

3. Clustering method was applied to segment the defect cluster from the remaining noise

And an example of the algorithm flow can be illustrated as:
![Fig 1-2 Algorithm visualization](https://github.com/zhouxf53/k-means-clustering-defect-inspection/blob/master/2.png)

## Usage
### Thermography images to temperature matrix conversion
This repository contains one folder called “T2” which is the test images collected in our experiment. Those raw images were stored in .tiff format and can be converted to 160*120 temperature matrix in .txt. format using the Visual Basic program called ‘’read_tiff_new.zip’’ inside of the same repository. So that each image would be a matrix of 160*120*1, and every image inside of the folder can be subject to further image processing as an input of 3D matrix of 160*120*n.

### Execution of the code
The code was written in both MATLAB and python, pick **either one** you like. Before you run it, double check the filename/folder name line to make sure it is corresponded to the current folder you want to process. The algorithm should generate a binary image containing only the defect area in the end

### Acknowledgements
Dr. Hongjin Wang and Bo Peng contributed a lot to the submitted algorithm

### Citation
If this method is useful for your research, please cite our paper.

### To-do
Cluster number justification, python code revision, matlab code comment adding