本篇論文主要針對TFT-LCD玻璃基板表面做自動化瑕疵檢測，並且發展瑕疵自動分類技術。架構分成硬體影像擷取裝置以及軟體影像處理技術。影像擷取方式選擇Line-scan模式，實際調整光源與攝影機的相對角度可以凸顯TFT-LCD玻璃基板的瑕疵。在影像前處理技術上，使用Gaussian Smoothing移除雜訊，以及Histogram Equalization改善panel間光線不均勻的問題。由於某些瑕疵非常細微難辨，因此利用2-D Wavelet Transformation強化瑕疵區域，增加瑕疵與非瑕疵的差異。在瑕疵自動檢測方面，使用Histogram Threshold、Difference of Block Mean以及Sobel Edge三種檢測方法，並設計實驗評估四種方法的優劣。為了進一步協助廠商對於受損的TFT-LCD進行修復處理，將依照瑕疵的特性自動分類以及量化損傷程度。

We present a inspection system for patterned TFT-LCD panel surface. In addition, we also present a technique for defect classification. There are two major parts in our framework, image acquisition and image processing. What type of camera we have decided is Line-scan camera. Mandatory of a successful inspection system is the correct selection of relative angle between camera and illumination, which needs to reveal the defects to be detected. The preprocessing stage uses Gaussian Smoothing to remove noise and Histogram Equalization to obtain a uniform lighting. Because some defects are too slight to recognize, using 2-D Wavelet Transformation to enhance defects. In automatic defect detecting stage, presenting three methods, Histogram Threshold、Difference of Block Mean and Sobel Edge. According to the experiment, we decided the best one among methods. In order to help the follow-up movement for defected panel, we classify defects and decide damaging degree according to the characteristic of defects.

none