OLED panel defect detection using local inlier-outlier ratios and modified LBP

We present an automated system for detecting surface defects on OLED panels. These panels exhibit varying textures and patterns which complicates the defect detection process. These detection systems have to be highly accurate and reliable as even a small error in detection can cause huge losses. In this paper, we present a method for detection of OLED panel surface defects using a novel and simple set of features based on local inlier-outlier ratios and modified LBP. The proposed inlier-outlier vector is easy to compute and provides robust discrimination between defect and non-defect samples of micro defects such as scratches and spots which are missed by modified LBP, thus proving to be a good complement to the modified LBP vector. Next, we train a SVM classifier using the concatenation of inlier-outlier ratios and modified LBP features. In the experiments, we have evaluated our method on several defects like scratch, spot, stain and pit, and the results show that our method significantly outperforms methods which use only modified LBP approach with minimal increase in computational complexity.

1 Introduction

Due to various advantages of OLED displays like no motion lag, perfect viewing angle, razor sharp display of images, excellent brightness levels and low power consumption, they have become a seemingly ubiquitous part of our daily lives in the form of mobiles, televisions, monitors, smart watches, etc. However, a major disadvantage of these displays is their complex manufacturing process which results in a variety of surface defects on the panels. It is important that only high quality panels get shipped as a part of the product to ensure customer satisfaction. Hence, automated inspection of the panels becomes an important part of the manufacturing and assembly process.