Regularized Auto-Encoder-Based Separation of Defects from Backgrounds for Inspecting Display Devices

Abstract

We investigated a novel method for separating defects from the background for inspecting display devices. Separation of defects has important applications such as determining whether the detected defects are truly defective and the quantification of the degree of defectiveness. Although many studies on estimating patterned background have been conducted, the existing studies are mainly based on the approach of approximation by low-rank matrices. Because the conventional methods face problems such as imperfect reconstruction and difficulty of selecting the bases for low-rank approximation, we have studied a deep-learning-based foreground reconstruction method that is based on the auto-encoder structure with a regression layer for the output. In the experimental studies carried out using mobile display panels, the proposed method showed significantly improved performance compared to the existing singular value decomposition method. We believe that the proposed method could be useful not only for inspecting display devices but also for many applications that involve the detection of defects in the presence of a textured background.

A Review and Analysis of Automatic Optical Inspection and Quality Monitoring Methods in Electronics Industry

“…This method is suitable for detecting defects of highly periodic textured surfaces; however, the variety of LCD manufacturing processes and the higher image resolutions of LCD products often lead to less periodical, more complicated textural structures in sensed images [245]. Furthermore, the proposed method is highly sensitive to texture direction and face difficulties in separating faint and large defects such as Mura from the texture background [242], [286]. Kim et al in [99] used Regularized Singular Value Decomposition (RSVD) method, which is based on SVD, as a feature extraction approach to extract failure patterns of DRAM WBM.…”

“…Furthermore, compared with MSCDAE used in [250], MS-FCAE requires no redundant computation at all. Jo and Kim in [286] proposed a regularized auto-encoder algorithm as a features extraction approach to separate the background of images from the defective regions for inspecting displays of mobile devices. This study focused on comparing their approach to similar feature extraction and dimension reduction approaches low-rank-approximation-based separation methods (specifically SVD) and the segmentation-based method.…”

72‐4: Invited Paper: Synthetic Defect Generation for Display Front‐of‐Screen Quality Inspection: A Survey

“…Those rules usually come from industrial use cases and are extracted by experienced engineers. For example, Jo et al [6] generated five different types of display defects (vivid dot, faint dot, line, stain, and mura) by specifying a set of rules that control the shape and appearance of defects on each sample. The generated defective samples are shown in Figure 1.…”

“…with improved quality at a decreasing cost. All of these factors pose challenges for quality control of the display manufacturing process, where image-based inspection is an essential task [6,7]. To achieve accurate defect detection for display panels, deep learning can be used [8].…”

A Mura Detection Model Based on Unsupervised Adversarial Learning

“…In the field of Mura detection, supervised deep learning has been carried out. Heeyeon Jo et al [28] used an automatic encoder to separate defects. Hua Yang et al [29]proposed an online sequential classifier and transfer learning (OSC-TL) method for online training and classification of Mura defects.…”