應用實驗設計於LCD面板自動化光學瑕疵檢測之影像處理方法選擇

Abstract

薄膜電晶體液晶顯示器(TFT-LCD)是電子產品上重要的介面顯示器，各廠商之技術日趨成熟，在技術相當的競爭下，維持良好品質並減少不必要的浪費是領先的一大關鍵，產品進行百分之百全檢也是必然的趨勢，然而，有部分廠商對於液晶顯示器上瑕疵的檢測，仍藉由人工目視的方式進行，此種方式將會耗費大量人力成本，且檢驗時間長，檢測作業人員容易受到其他因素影響，因此如何以自動化技術提高產品品質是產業所面臨的一大課題。自動化光學檢測(Automatic Optical Inspection, AOI)，是高速度、高精確度的光學影像檢測系統，運用「機器視覺」做為檢測技術，再搭配視覺感測設備，檢測出產品中的瑕疵，判斷並挑選出產品，或是用於量測產品尺寸等，而目前已有多數產業導入自動化光學檢測於各項產品上，為了提高最後檢測時的檢出效果，檢測流程中一定會加入影像處理之步驟，因此影像處理技術是整個AOI的重要環節之一，影像處理技術有許多種類，每種技術的處理效果也都不同，而執行越多的強化影像的動作絕對能提升檢出率，但在時間上勢必也會增加，故檢出率與檢測時間上的取捨也是需要考量的重點，本研究欲使用實驗設計，分析在LCD面板的檢測流程中，何種影像處理的組合能提供最佳的檢測效率。首先，本研究根據G公司所提供的真實瑕疵圖進行瑕疵圖製作，接著透過依實驗組合編寫的影像處理程式取得各類瑕疵之檢出率與檢測時間等數據，最後再進行相關分析，由實驗結果可得知，使用中值濾波加上銳化處理的影像處理程序在各類瑕疵上皆具有最高之檢出率，但在檢測時間的考量下，本研究會提出使用高斯濾波加上銳化處理的影像處理程序作為LCD面板AOI檢測流程之參考。

Automatic Optical Inspection(AOI) is a high speed and high accuracy detective system, which using machine vision as a technique, then detecting defect or measuring size to determine and choose the product. For improving the efficiency of detection, the procedure of detection would include one or more image processing techniques. Therefore, the image processing which the step in detection procedure is one of the important part in Automatic Optical Inspection. There are many kinds of techniques in image processing, each kind of technique has its own effect. Implement more techniques would have higher detectable rate, but that needs more times to detect. How to trade off between detectable rate and detection times is also this research focusing on. This research would applies design of experiment to analyze which combination of image processing techniques in detection procedure for TFT-LCD panel could have the best detection efficiency. First, in order to increase sample size, this research will use the real defect image which G company provided to make the simulated defect image. Then according to experiment treatment to code the image processing program to collect detectable rate and detection time of each defect. Lastly, analyze the data by statistical analysis. According the result of experiment, using median filter and image sharpening in the detection procedure have the highest detectable rate, but it will spend the most times. Consequently, considering the detection times, this research would suggest using Gaussian filter and image sharpening in the detection procedure for the reference of Automatic Optical Inspection of TFT-LCD panel.