一維機器視覺技術於複雜結構之TFT-LCD面板表面瑕疵檢測

Automatic Surface Inspection for TFT-LCD Panels with Complicated Patterns Using 1D Machine Vision

莊樹達(Su-ta Chuang)

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(2004年)

摘要

薄膜電晶體液晶顯示器（Thin film transistor liquid crystal display, TFT-LCD）輕薄與可攜帶性帶給人們生活上更多的便利，而多象限垂直配向型（Multi-Domain Vertical Alignment, MVA）TFT-LCD克服了過去LCD視角狹窄之缺點，而成為顯示器市場的主流。現今TFT-LCD製造技術朝向解析度越高、尺寸大的發展趨勢，在元件構造上也更加的精細及複雜，MVA-LCD面板除了呈現水平與垂直排列的信號線（Data line）、閘線（Gate line）之外，更包含了結構複雜的薄膜電晶體（TFT）與共通線（Common line）等元件，其面板紋路呈現週期排列特徵，而現有無參考樣本之檢測方法無法延伸於MVA-LCD面板瑕疵檢測，所以本研究導入一維機器視覺技術於MVA-LCD面板表面瑕疵檢測，檢測對象為人工不易檢測之微觀瑕疵，如粉塵（Particle）、水漬（Watermark）、纖維（Fiber）等。

本研究模擬線性一維掃描方式，針對複雜結構紋路之MVA-LCD影像的微觀瑕疵，發展出二個不需要參考影像之瑕疵檢測的機器視覺演算法，藉由擷取之一維灰階影像呈現週期性變動之特徵，分別以空間域及頻率域兩種方法進行瑕疵檢測。空間域檢測方法中，令影像訊號經碎形轉換得到更突顯週期性之一維碎形訊號，再進行區段相關係數之運算，並設定臨界值判別出瑕疵區段；在頻率檢測方法中，則將切割之各區段重組成二維影像，以傅立葉轉換及還原技術，於傅立葉頻譜中去除規律性紋路所響應之頻率元素，最後經傅立葉還原後再以管制界限法突顯瑕疵達到瑕疵偵測的目的。兩種方法中，空間域檢測方法僅能判斷出瑕疵所存在之區段，而頻率域檢測方法則能檢測出瑕疵真正的位置及形狀，但在運算時間上，空間域方法低於頻率域方法達10倍以上。

在檢測結果上，運用空間域及頻率域檢測方法皆可以有效的檢測出MVA-LCD面板上各類型之微觀瑕疵，本研究也同時結合兩種方法進行兩階段之瑕疵檢測，不僅能降低運算時間，並能達到頻率域方法之檢測結果。

關鍵字

TFT-LCD ； 線性掃描 ； 表面瑕疵檢測 ； 相關係數法 ； 傅立葉轉換

並列摘要

Thin Film Transistor Liquid Crystal Displays (TFT-LCDs) have become increasingly attractive and popular as thin, light, and low power-consumption display devices. The main trends of TFT-LCD now are toward “Large Size” and “High Resolution”. The structural textures of newly developed Multi-domain Vertical Alignment (MVA) TFT-LCD panels are much more complicated than those of traditional models such as Twisted Nematic (TN) TFT-LCD. In this research, two one-dimensional defect inspection schemes are proposed to detect the micro defects in MVA TFT-LCD surfaces that contain complicated structural textures in images.

The proposed inspection methods do not rely on a template image for comparison. They are based on the detection of a periodic pattern in each MVA TFT-LCD line image at high resolution. A spatial-domain method and a spectral-domain method are developed in this research. In the spatial domain, we use the fractal transformation to enhance the periodicity of a 1D gray-level line image, and divide the fractal signal into small segments, each of the length of the repeated period of an MVA TFT-LCD panel. By calculating each segment’s normalized cross correlation (NCC) with its two neighboring segments and comparing the NCC value with a predetermined threshold, we can easily identify the segment that contains defects. In the spectral domain, the original 1D gray-level line image is divided into segments with the length of a period, and then the divided 1D segments are constructed as a 2D image. By eliminating the frequency components in the Fourier spectrum that represent the periodical structural pattern of the constructed 2D image, and back transforming the image using the inverse Fourier transform, it will effectively remove the repetitive background pattern and well preserve defects. The spatial method is computationally efficient, but it can only detect the segments that contain defects. The spectral method can identify the shape and location of a detected defect, but it takes more computational time, compared to the spatial method.

Experimental results have shown that the two proposed methods can well detect various defects in complicated TFT-LCD panel surfaces. In this study, a hybrid scheme that combines both the spatial and spectral methods is also proposed. It first uses the spatial method to efficiently identify the defective segments, and then applies the spectral method to the defective segment to effectively identify the shape and exact location of the defect in the segment. Experimental results have also verified the efficacy of the hybrid scheme for defect detection in MVA TFT-LCD panels.

並列關鍵字

TFT-LCD ； 線性掃描 ； 表面瑕疵檢測 ； 相關係數法 ； 傅立葉轉換

參考文獻

郭志嘉，2002年，「應用三維傅立葉於非同質性隨機性紋路表面之瑕疵檢測」，碩士論文，私立元智大學工業工程研究所。

曾彥馨，2003年，「應用機器視覺於TFT面板之表面瑕疵檢測與分類」，碩士論文，私立元智大學工業工程研究所。

謝志雲，1998年，「應用機器視覺於方向性紋路之表面瑕疵檢測」，碩士論文，私立元智大學工業工程研究所。

洪崇祐，2004年,「應用一維傅立葉分析於TFT-LCD液晶顯示面板之瑕疵檢測」，碩士論文，私立元智大學工業工程研究所。

Bodnarova, A., M. Bennamoum, and K. Kubik, 2000, “Suitability analysis of techniques for flaw detection in textiles using texture analysis,” Pattern Analysis and Applications, Vol. 3, pp. 254-226.