顯示器(Display)是資訊時代人們訊息傳遞與溝通之重要界面,平面顯示器(Flat-panel displays)具有輕薄與可攜性更加帶給人們許多生活上之便利,形成了繼矽品積體電路後,突飛猛進的新科技領域。目前TFT-LCD的瑕疵檢測仍然仰賴大量人工進行檢測。本研究導入機器視覺於TFT面板之表面瑕疵檢測,以避免人工檢測誤判造成之損失,有效降低生產成本與提昇良率。本研究瑕疵偵測對象設定為TFT面板中人眼不易觀察之微觀瑕疵,包括指觸、粉塵、配向膜孔洞與刮痕瑕疵,透過對TFT面板之瑕疵偵測於製程中瑕疵偵測出瑕疵位置,且進一步將此四類瑕疵加以分類,回饋生產線及時且快速之製程變異資訊。 本研究主要針對TFT面板先進行微觀瑕疵發生區域的偵測(Detection),再進行瑕疵的分類(Classification),由於TFT面板表面是由垂直與水平之規律紋路所構成,因此本研究針對瑕疵偵測部分利用傅立葉影像轉換(Fourier transform)與反傅立葉影像還原的技術,配合濾除頻譜中TFT面板垂直水平紋路之頻率元素與保留瑕疵能量在頻譜中所貢獻最多的區域之兩種策略,將TFT面板規律紋路加以濾除並且保留瑕疵特徵,再經過統計管制界線法凸顯瑕疵達到瑕疵偵測目的。透過瑕疵偵測後本研究方法接著對異常區域的瑕疵加以分類,在傅立葉還原影像中先利用霍夫轉換法分離刮痕,再利用面積特徵判別孔洞與粉塵,透過型態學之閉合處理群聚指紋為一區塊後利用面積特徵加以分辨。經實驗驗證本研究方法對TFT面板之微觀瑕疵的偵測皆具有良好效果。
The surface defects in TFT-LCD array panels generally can be categorized as macro- and micro-defects. Macro-defects are large in size and can be visually identified by human inspectors. However, sizes of micro-defects are generally very small and cannot be easily detected by human personnel. In this study, we propose an automatic visual system for defect detection and classification, and especially focus on micro-defects in TFT-LCD array panels in the manufacturing process. Micro-defects of TFT-LCD array panels investigated in this study include pinholes, particles, scratches and fingerprints. Since a TFT-LCD array panel comprises repetitive vertical and horizontal line patterns, it can be considered as a homogeneously structural texture. The proposed method does not rely on local features of textures. It is based on a global image reconstruction scheme using the 2D Fourier transform. The Fourier spectrum is ideally suited for describing the directionality and periodicity of line patterns in a TFT-LCD image. By eliminating the frequency components that represent the background texture of the TFT-LCD surface, and back transforming the image using the inverse Fourier transform, we can effectively remove the repetitive line patterns and enhance local defects in the reconstructed image. The geometric features of each detected defect in the reconstructed image are then extracted for further classification. Experimental results from a variety of TFT-LCD array samples have shown the efficacy of the proposed method for detecting and classifying micro-defects.