Pretest Gap mura on TFT LCDs using the optical interference pattern sensing method and neural network classification

Tung Yen Li, Jang Zern Tsai, Rong Seng Chang, Li Wei Ho, Ching Fu Yang

Research output: Contribution to journalArticlepeer-review

31 Scopus citations


Recently, thin-film transistor liquid crystal displays (TFT LCDs) have had a high demand in the market, which entails careful product quality control and more stringent defect detection procedures. A good defect detection rate is the basic requirement of the quality control process. The use of conventional human visual inspection methods to find the defects in TFT LCDs is simply not accurate enough and consumes a large amount of resources. An automatic defect inspection method is thus necessary for this industry; to find the defects, the type of defects needs to be recognized as well. Here, we propose an inspection procedure based on the optical interference pattern sensing method to find the interference fringes and then use the image processing to enhance the contrast of the interference fringes, thereby increasing the recognition rate for the latter process. The neural network method is used to learn about and identify the defects and their types. This paper focuses on the mura defect inspection and classification method. Before the learning process has begun, the mean squared error was roughly three, but after neural network retraining of these samples, the results showed that the mean squared error was less than 0.01. The defective panels can be sorted out using this method so that the next processing and waste of materials can be avoided.

Original languageEnglish
Article number6236141
Pages (from-to)3976-3982
Number of pages7
JournalIEEE Transactions on Industrial Electronics
Issue number9
StatePublished - 2013


  • Image process
  • interference
  • mura
  • neural network
  • thin-film transistor (TFT) liquid crystal (LC) display (LCD) (TFT LCD)


Dive into the research topics of 'Pretest Gap mura on TFT LCDs using the optical interference pattern sensing method and neural network classification'. Together they form a unique fingerprint.

Cite this