16‐4: Invited Paper: Region‐Based Machine Learning for OLED Mura Defects Detection
Author:
- 1 May 2021
- journal article
- research article
- Published by Wiley in SID Symposium Digest of Technical Papers
- Vol. 52 (1) , 200-203
Abstract
Automatic inspection of Mura defects is a challenging task in display manufacturing. Due to the nature of Mura defects, which appear as brightness variances in the low‐contrast images captured by the optical inspection camera, the defects are extremely difficult to detect because they show no clear edges from their surroundings while the image backgrounds usually present uneven illumination. In this paper, I propose an effective way to detect two types of Mura defects using a region‐based machine learning approach. My research includes three components: 1) creating a quality dataset from the raw optical inspection images, 2) designing a region‐based machine learning model with a preprocessor, a candidate detector, a feature extractor, and a classifier, and 3) an adversarial training and evaluation method to overcome the inconsistent data labels. Applying real panel images from the display manufacturing line as my test set, my trained model achieved a recall rate of 100% and a precision rate of over 90%.Keywords
This publication has 3 references indexed in Scilit:
- Ambiguous Surface Defect Image Classification of AMOLED Displays in SmartphonesIEEE Transactions on Industrial Informatics, 2016
- GPU Architecture for Unsupervised Surface Inspection Using Multi-scale Texture AnalysisProcedia Technology, 2014
- Selective Search for Object RecognitionInternational Journal of Computer Vision, 2013