loading page

Ensemble Based Automotive Paint Surface Defect Detection Augmented By Order Statistics Filtering Using Machine Learning
  • +3
  • Adnan Shahid Khan,
  • Muhammad Hasan Mehdi,
  • Riaz Uddin,
  • Abdul Rehman Abbasi,
  • * BSChowdhry,
  • Kashif Nisar
Adnan Shahid Khan
Universiti Malaysia Sarawak Fakulti Sains Komputer dan Teknologi Maklumat

Corresponding Author:[email protected]

Author Profile
Muhammad Hasan Mehdi
NED University of Engineering & Technology
Author Profile
Riaz Uddin
NED University of Engineering & Technology
Author Profile
Abdul Rehman Abbasi
Pakistan Institute of Engineering & Applied Sciences
Author Profile
* BSChowdhry
Mehran University of Engineering & Technology
Author Profile
Kashif Nisar
Swinburne University of Technology
Author Profile


In automotive industry, the paint surface inspection process is still mainly performed by manual methods based on subjective human vision system, which may not only be inaccurate but time consuming and costly as well. Recent promises by machine vision, image processing and machine learning, techniques have led to emergence of tools may now allow development of robust models that may be successfully used to perform automatic paint surface inspection. Hence, modeling of such a system for the purpose of automation provides opportunity in reducing the cost and time attributed to inspection and repairs. This paper reports a novel study towards an effective detection and classification of various surface defects attributed to different color mixes or paint textures. Proposed method performs the detection using the combination of non-linear spatial order statistics filter with Gaussian filter for preprocessing followed by Canny Edge Detection and Morphological Transformation. Next feature extraction and classification using a voting classifier selected among eight different classifiers and their performance analysis. Results show that our model performs defect detection with high accuracy, precision and recall (91.17%, 91.14 %, 91.17%).