本研究主要探討出現於薄膜電晶體液晶顯示器(Thin film transistor liquid crystal display, TFT-LCD)上的Mura瑕疵,此種瑕疵主要會造成LCD面板在顯示畫面時,會呈現偏亮或偏暗之區塊;而Mura瑕疵可能發生於LCD製程中的各區段,該瑕疵有亮度不均以及低對比的特性,其特性會造成Mura瑕疵在二維影像上的檢測困難性,但是由LCD影像之水平或垂直方向的灰階掃描線則可顯示正常一維訊號與異常一維訊號之差異,因此本研究將利用機器視覺之技術,針對影像中水平及垂直方向,以一維線性掃描的方式進行瑕疵檢測。 本研究提出二階段檢測方式進行Mura瑕疵檢測,其中第一檢測階段利用獨立成份分析(Independent Component Analysis, ICA)設計能代表正常一維訊號之特徵濾波器,此濾波器對於正常一維訊號會有較為一致的濾波反應值,而異常一維訊號則會有變異較大之濾波反應值,透過濾波反應值來判斷一維訊號是否發生異常;第二檢測階段利用小波轉換(Wavelet transform)能突顯細部變異的能力,針對第一檢測階段中判斷為異常之一維訊號,進行小波拆解並得到細節訊號,透過細節訊號進行瑕疵位置的偵測,並將Mura瑕疵位置標示出來。 本研究實驗對象包含LCD面板以及彩色濾光片的Mura影像(影像大小皆為 像素),在檢測時間上,根據現有個人電腦之計算,正常影像約需0.1秒,Mura影像約需0.2~0.3秒可完成檢驗,可有效降低檢測所需之時間;而檢測的Mura瑕疵包含Spot mura、Ring mura、Gravity mura以及Weak-line mura,由檢測結果可知,二階段檢測法皆可有效判斷Mura瑕疵,並偵測瑕疵之位置。
This research proposes a machine vision method for detecting non-uniform brightness defects, called mura, in TFT-LCD manufacturing. Mura will cause bright or black blocks when LCD panels are switched on. It has low-contrast and uneven brightness properties, and causes inspection difficult in 2D images. By horizontally and vertically scanning the 2D image of an LCD surface, the 1D gray-level profiles between faultless and defective line images can be well distinguished. One-dimensional inspection schemes are, therefore, proposed for mura detection in low-contrast images. The proposed method in this study involves two detection stages. Independent component analysis (ICA) is adopted first to design filters that can best represent the global features of faultless line images. The first detection stage uses the convolution filters to identify suspected line images. In the second detection stage, the potential defect line images are evaluated, and the true defect location is detected using the high-pass Haar wavelet decomposition. The proposed method involves only convolution operations in the two detection stage, and is very computationally efficient. The experimental results have shown the efficacy of the proposed method for a number of LCD panels and color filters that contain various defect sizes and shapes including spot-, line-, ring- and gravity-mura.