TFT-LCD Mura defects automatic inspection system using linear regression diagnostic model

Abstract

The TFT-LCD panel is one of the most important and promising products in the recent years. Mura defects can be created on the display panel during its production. In this research, a linear regression diagnostic model is incorporated with digital image processing theory to automatically inspect for Mura defects. A bivariate polynomial regression model is used to simulate the brightness of background images that is used in the diagnosis of outliers and influential points. The partitions of the candidate Mura defect regions are segmented using Niblack's threshold criteria. The candidate Mura defects are further evaluated. The quantified level is defined in terms of concepts already reported in the literature. Based on Weber's law and a visual perception model, the just-noticeable intensity difference index of the Mura features can be obtained and it can be subsequently used to quantify the Mura defect level. With the obtained defect level, Mura defects can be identified for exactly labelling of the perfect and imperfect LCD panels.Experiments were performed on 13 TFT-LCD panels. There are ten bad panels and three good panels in these 13 samples as determined by human visual inspection. Each bad panel has at least one Mura defect. After the automated inspection process, the results showed that the proposed method could separate the good and bad panels accurately. Compared with human visual inspection, the Mura detection rate of the distinct size and shapes can attain over 90.9 per cent correct detection and the achieved correction rate of Mura defects on each panel can be improved by 100 per cent.

A new algorithm on the automatic TFT‐LCD mura defects inspection based on an effective background reconstruction

“…The detected result obtained through image subtraction is further evaluated using the SEMI formula with the JND definition for mura quantification. The SEMI formula for mura quantification (SEMU) value can be defined as follows: where C x is the average contrast; C JND is the JND of mura defect, and A is the area of mura defect (pixels). The candidate region‐mura is identified to be real if its level exceeds the standard value defined by the industry.…”

“…This study focuses on the development of a method that accurately reconstructs the background from the DUT image. The background reconstruction is based on the following three well‐known methods: polynomial surface fitting, DCT, and low‐pass filtering. The main objective can be achieved by conducting the following steps: Developing an algorithm to adaptively reconstruct the background image with high accuracy Applying the reconstructed background to the image segmentation process based on the sensitivity of the human eye Developing an efficient computation process Validating the proposed method by comparing the results obtained with those of previous studies. …”

“…5 Based on the sensitivity of the human eye to mura, the level for each mura candidate is quantified using the concept of just-noticeable difference (JND), which is used to identify real muras by grading as either pass or fail. Two methodologies for the detection of mura defects have recently been proposed: background image reconstruction [5][6][7][8][9][10][11][12][13][14][15][16] and image segmentation technique. [16][17][18] The reference image can be obtained by background image reconstruction of the DUT source image or by using the representation of basis images.…”

“…This study focuses on the development of a method that accurately reconstructs the background from the DUT image. The background reconstruction is based on the following three well-known methods: polynomial surface fitting, DCT, 11 and low-pass filtering. The main objective can be achieved by conducting the following steps:…”

Automatic TFT-LCD mura detection based on image reconstruction and processing

“…However, there are always visual quality defects generating in the manufacturing process of TFT-LCD due to the complex steps, among which mura defects are the most critical ones and mostly difficult to be successfully detected in the human-laboring detection process. To solve this problem, many researchers have proposed varied methods trying to inspect it automatically [2], [3], [4], [5]. Nevertheless, it remains to be a challenging task in TFT-LCD quality control steps due to the following characteristics held by mura defects: (1) Mura has no feature representations that can be accurately quantified.…”