一維經驗模態分解法於TFT-LCD面板影像之 Mura (光源不均)瑕疵檢測

Low-contrast surface inspection of mura defects in TFT-LCDs using Empirical Mode Decomposition

吳秋燕(Chiu-Yen Wu)

指導教授 ： 蔡篤銘

元智大學/工程學院/工業工程與管理學系/碩士(2007年)

https://doi.org/10.6838/YZU.2007.00136

摘要

TFT-LCD製程中有一種面板瑕疵稱之為Mura，該瑕疵是指顯示器亮度不均勻造成各種痕跡的現象，此類瑕疵存在於光源不均的背景上，人類肉眼無法有效地將正常影像與Mura瑕疵分辨出來，現階段在製程上仍然依賴作業員利用目視的方式進行檢測。由於Mura在二維影像中與背景對比度低，縱使利用影像強化的方式突顯Mura瑕疵，卻也反映了影像光不均的現象；若觀察TFT-LCD面板之一維影像，正常的一維影像灰階會呈現一直線的趨勢，而具Mura瑕疵的一維影像灰階訊號，其瑕疵區段趨勢則具有較明顯的波動，因此本研究利用一維影像訊號來進行TFT-LCD面板影像之Mura瑕疵檢測。

本研究開發一個兩階段之機器視覺檢測方法：第一檢測階段利用線性逼近法偵測出異常一維訊號；再將第一檢測階段判為異常之一維訊號採用經驗模態分解法(EMD)分解的本質模態函數與殘差，由此訊號中判別出瑕疵區段。由於EMD能夠將原始訊號分解出多個本質模態函數及一個殘差，每個本質模態函數(IMF)會對應原始訊號中的單一頻率，殘差代表原始訊號的整體趨勢，因此利用IMF及殘差找出Mura瑕疵的位置。本研究針對19張LCD面板影像進行實驗，在這些待測影像內，正常影像都能正確判斷為無瑕疵影像，而Mura瑕疵影像(包括Gravity、Spot、Ring 及Weak-line mura)在瑕疵形狀上雖有些不完整，卻能夠將瑕疵位置檢測出來，目前本研究方法在應用上仍受限於EMD分解之計算時間。

關鍵字

機器視覺 ； 瑕疵檢測 ； 液晶面板 ； 經驗模態分解法 ； Mura

並列摘要

There is a large category of defects, called mura, in TFT-LCD manufacturing. Mura appears as a low-contrast and non-uniform brightness region in the image. The human inspectors are unable to distinguish between the normal regions and mura effectively. A mura defect in the two-dimensional image has low contrast to the background, and shows no clear edges from its surrounding neighborhood. An enhanced image of the LCD surface only intensifies the intrinsic non-uniformity of the LCD panel, which makes the detection task even more difficult. When mura is scanned horizontally or vertically in the image, the gray-level profile of a faultless 1D image appears as a straight line, and the defective 1D image of mura has a distinct oscillation. Therefore, this study proposes a one-dimensional machine vision scheme to detect defects in low-contrast TFT-LCD panel images.

The proposed method in this research involves two detection stages. The first detection stage utilizes a fast line fitting to detect anomalous 1D signal. Then the suspected defect signal is further verified, and the location of the defect is extracted in the second detection stage using empirical mode decomposition (EMD). EMD is used to generate a collection of intrinsic mode functions (IMF) and a residue. The IMF in each cycle involves only one mode of oscillation of the input signal. It is locally mono-frequent. The residue represents the global trend of the primitive signal. The second detection stage utilizes IMFs and the residue to identify the location of mura defects. In this research, 19 true LCD panel images have been evaluated. The experimental results show that the proposed method can effectively detect various mura defects including spot-, gravity-, ring- and line-mura, yet no defects are declared for all faultless LCD images. The proposed method in its present form is very computationally fast in the first detection stage. However, it is computationally intensive in the second stage due to the sifting process of EMD.

並列關鍵字

Machine vision ； Defect detection ； TFT-LCD panels ； Empirical Mode Decomposition ； mura

參考文獻

蘇世斌、沈駿緯、郭淑純、郭浩然、呂志維，2004年，「TFT-LCD廠點燈檢測作業人員之眼部自覺症狀調查」，中華職業醫學雜誌，11卷2期，pp.107-115。

陳文杰，2006年，「雜訊環境下經驗模態分解法於語音辨識之應用」，碩士論文，國立中央大學電機工程研究所。

Echeverria J C, Crowe J A, Woolfson M S and Hayes-Gill B R., 2001, ”Application of empirical mode decomposition to heart rate variability analysis.” Med. Biol. Eng. Comput., Vol. 39, pp. 471-479.

G. Rilling, P. Flandrin, and P. Goncalves, 2003, “On empirical mode decomposition

and its algorithms,” Proc. IEEE-EURASIP Workshop on Nonlinear Signal and Image Processing.