透過您的圖書館登入
IP:3.133.87.156
  • 學位論文

應用機器視覺技術於雙穩態膽固醇液晶顯示面板之瑕疵檢測

Apply Machine Vision Techniques to Defect Detection for Bi-stable Cholesteric Liquid Crystal Displays

指導教授 : 劉益宏

摘要


雙穩態膽固醇液晶顯示器(Bi-stable Cholesteric Liquid Crystal Display, BS-ChLCD)是一種具有輕、薄、高對比、可撓曲、省電、具記憶性等特點的新型顯示技術。近年來類紙式(Paper-Like)的軟性顯示器,舉凡書報、標籤與市面上的消費型顯示器等,除了將帶來更人性化與行動化的便利之外,同時也會大幅改變現有生活型態。但以產品開發的角度看來,關鍵性的元件與材料尚處於研發階段,以致於面板生產製作過程中,將無法避免人為或機台等因素而產生瑕疵,進而影響到面板的顯示效果。因此,本論文導入機器視覺於膽固醇液晶顯示面板進行表面瑕疵檢測,此系統主要由三個程序所構成:影像前處理、訓練程序及檢測程序。本研究基於支持向量機器(Support Vector Machine, SVM)所建構之瑕疵檢測方法是利用一張原始影像切割成數張子影像,接著進行瑕疵分類器的訓練。但待測影像並不是呈現完全水平的狀態,而這樣的誤差將導致後續重建背景紋路上的困難,因此在進入訓練及檢測程序之前,必須先將影像調校為水平,以利於後續檢測流程。在訓練程序中,利用紋理特徵(Texture)、變異數(Variance)、奇異値分解(Singular Value Decomposition, SVD)及主成分分析(Principal Component Analysis, PCA)等特徵抽取方法,亦即每張分割後的子影像經過特徵抽取便形成一組特徵向量,最後依據SVM理論,利用這些特徵向量建構出一個瑕疵檢測模型。檢測程序中,將一張原始影像經過前處理後,以相同的特徵抽取方式輸入至瑕疵檢測模型,便可判斷該張子影像是否具有瑕疵,若存在瑕疵,則將這張子影像加以標記;若無瑕疵,子影像進入背景紋路的斷線偵測系統,直到所有子影像都經過一系列的瑕疵判別之後,即完成一張影像的檢測。透過實驗的結果,本論文所提出的瑕疵檢測方法,平均分類率可達99.04%。

並列摘要


Bi-stable cholesteric liquid crystal display (BS-ChLCD) is a novel display technology with characteristic of light, high contrast, bendable, energy saving, and memorability. Recently, a consumer revolution in people’s way of reading is caused by Paper-Like flexible display with portable and mobility. However, the key components and materials have yet been exploited really perfect. Furthermore, the performance of displays would be affected by the production imperfections. Therefore, in the research, the Machine Vision System would be applied to inspect the exterior defect with three mechanisms: Vision Pre-processing, Training Procedure, and Inspecting Procedure. The proposed system is based on Support Vector Machine (SVM) to decompose an original image to several sub-images then processing the defect classified mechanisms. The original images are not full horizontal; it is very difficult to reconstruct the background patterns with this kind of horizontal error. Nonetheless, the Vision Pre-processing systems have to tune the original images to full horizontal for following inspection procedure. The Training Procedure mechanism is construction of Texture, Variance, Singular Value Decomposition (SVD), and Principal Component Analysis (PCA), according to the analysis results, each sub-images are with unique eigenvector. The proposed systems build a defect inspection model via those eigenvectors based on SVM. In this model, each original image would be split up into sub-images and input the eigenvector set calculated from those sub-images for defect judgments. Proceed to the next step, remark the sub-images if the defect was found. After processing all sub-images, the model would show the inspection result of defect judgments. According to the experiment result showed in this work, the recognition rate of proposed system is 99.04%.

參考文獻


[6] 陳志忠,“液晶顯示器的像素點缺陷與相對亮度均一性之自動化檢測,” 私立中原大學機械工程學系碩士論文,2001.
[7] 郭家成,“薄膜電晶體液晶顯示器Mura瑕疵檢測技術之研發,” 國立台北科技大學自動化科技研究所碩士論文,2006.
[14] 陳振昌,“軟性顯示器之光電特性量測,” 國立清華大學光電工程研究所碩士論文,2006.
[18] 王建達,“基板上預先形成之聚合物穩定反射式雙穩態膽固醇液晶顯示器之研究,”私立崑山科技大學電子研究所碩士論文,2005.
[19] 蕭清中,“基於色彩空間之軟性顯示器色彩均勻性分析-以膽固醇液晶顯示器為例,” 私立中原大學機械工程學系碩士論文,2009.

被引用紀錄


彭彥博(2015)。應用支持向量機器於液晶顯示器面板瑕疵分類-以G公司為例〔碩士論文,國立交通大學〕。華藝線上圖書館。https://doi.org/10.6842/NCTU.2015.00374
高永樺(2013)。情緒腦波分析及腦機介面之開發〔碩士論文,中原大學〕。華藝線上圖書館。https://doi.org/10.6840/cycu201300950
陳衍均(2012)。應用支持向量資料描述於LED瑕疵檢測〔碩士論文,中原大學〕。華藝線上圖書館。https://doi.org/10.6840/cycu201200721
王治翰(2012)。情緒腦波誘發範例建立及辨識〔碩士論文,中原大學〕。華藝線上圖書館。https://doi.org/10.6840/cycu201200692
潘彥霖(2010)。同步腦機介面特徵分析與研究〔碩士論文,中原大學〕。華藝線上圖書館。https://doi.org/10.6840/cycu201000711

延伸閱讀