在TFT-LCD面板中，彩色濾光片是重要的零組件之一。然而，目前在彩色濾光片的瑕疵檢測上仍是以人工檢查為主。人工檢查不僅耗費人力，而且在瑕疵的判斷上不夠客觀，檢測人員也可能因為疲勞而影響檢測結果。為了確保彩色濾光片的生產品質，就必須導入機器視覺檢測技術來提昇檢測效率。 由於彩色濾光片具有結構性紋路，其紋路主要呈水平方向和垂直方向分布，此結構性紋路將造成瑕疵檢測上的困難。本研究利用傅立葉轉換方式，在傅立葉頻譜中濾除代表水平及垂直紋路的響應，並搭配高斯低通濾波器來保留瑕疵響應貢獻最多的區域，達到濾除結構性紋路的目的。在低通濾波器截止頻率的選擇上，過去總是由人為調整，經比較後，以最符合人眼視覺者選定為最佳截止頻率。本研究透過觀察功率頻譜上的能量分布特性，由系統自動決定截止頻率。除此之外，針對偏轉影像提出修正方法，在影像定位不足的情況下，依舊能得到正確的檢測結果。 Color filter is one of the important components in TFT-LCD. At present, the defect inspection of color filter is always done by artificial inspection. However, the decision of color filter defect strongly depends on subjective judgment of the inspector. Since the difference of the subjective cognition among these inspectors, it is difficult to achieve the uniform inspection quality. Besides, inspectors may affect the inspection result due to languid sense. Therefore, in order to guarantee the output quality of color filter panel, it is necessary to introduce the machine vision technique which can raise inspection efficiency. Since the color filter panel involves regular grid texture which consist of vertical lines and horizontal lines, its defects will be inspected with difficulty. For this reason, we use Fourier transform to filter these components which stand for spatial line patterns in frequency spectrum. In addition, we match Gaussian low-pass filter to retain the largest number of the flaws response to achieve our objective. In the past on the selection of low-pass filter, most of the studies through artificial adjustment and comparison to determine better cutoff frequency range of low-pass filter. In this thesis, we inspect the energy distribution in power spectrum and decide cutoff frequency automatically from distribution characteristic. Furthermore, we propose a method to calibrate the image deflection and reduce the wrong interpretation of results.
