隨著科技的進步，液晶顯示器已成為現今顯示器的主流。顯示器的影像品質為消費者購買的指標之一，為確保顯示影像品質，瑕疵檢測甚為重要。而今，在成品的瑕疵檢測上，大多使用人工目視來作檢測，這容易造成瑕疵漏檢與品質不一的情況，其中以Mura(顯示不均)瑕疵最難被人眼查覺出來。Mura瑕疵是指顯示器亮度不均勻造成各種的痕跡，由於瑕疵現象不明顯，因此費時且容易發生漏檢及誤判的情形。針對上述種種的缺點，本研究發展出一套檢測系統來取代人眼。 在Mura瑕疵影像中，背景不均的現象使Mura瑕疵難以用一般閥值分離出來，因此本研究朝估測背景的方法，希望成功估測出背景，去除背景因素的影響，突顯Mura瑕疵。奇異值矩陣分解法是將原影像展開成數個基底影像，這些基底影像是由特徵值與相對應的特徵向量所構成，因此越大特徵值代表原影像越重要的特徵。背景不均屬全區域現象，為影像中主要成份，所以將最大特徵值還原出的基圖來代表估測背景。經由實驗發現，由於實際樣本的背景更加複雜，必需以分區方式來進行檢測。本研究亦提供一個最佳分區法則來解決上述問題，最後再配合國際半導體設備和材料協會(SEMI)依人因實驗所統計出的SEMU值，辨識真實的Mura。在軟體方面，利用C語言完成一套Mura自動檢測系統的人機介面，提供瑕疵的相關資訊。 With the development of technology, LCD(liquid crystal display) has become the core of displays nowadays. The quality of LCD is one of the targets making customers to purchase. To make sure the quality of LCD, defect detection plays an important role in this field. Nowadays, most defect detections made by human eyes often causes failure in defect detections and inconsistency of quality. Among these defects, MURA defection is most hardly to be detected. MURA defection result from the um-uniform brightness which makes traces on LCD. Due to the unapparent defection, it often takes lots of time to detect but often out of judgement. To solve these defections above, this research designs a set of detection system to replace human eyes. On the image of MURA detection, it’s hard to use normal threshold to detect MURA due to the brightness um-uniform of background. For this reason, this research intends to estimate background of LCD. With this method, we hope to eliminate the effect caused by the background to reveal MURA detection. Singular value decomposition (SVD) expands original image into several based images－these based images are composed of eigenvalue and responded eigenvector. Therefore, the bigger eigenvalue is, the more important feature the original image is. The um-uniform of background is major component, so we take the biggest eigenvalue and responded eigenvector representing major element — the background. Actually, from this experiment we know that the background of the samples is more complex, so we must to separate several blocks to achieve detection of goal. For above reasons, we provide an optimum method to overcome the problem and add the value of SEMU (which is an ergonomics experiment completed by Semiconductor Equipment Materials International, SEMI) to identify real Mura. In software, we complete an interface of automatic system of detection Mura which can provide related information of defect.
