偏光膜瑕疵檢測分類之研究

Study on Classification of Defect Inspection for Polarizing Films

張仲華(Chung-Hua Chang)

指導教授 ： 郭文嘉

元智大學/資訊學院/資訊管理學系/碩士(2009年)

摘要

本論文提出一套偏光膜（Polarizing Film）自動化光學檢測（Automatic Optical Inspection , AOI）系統，主要應用影像處理技術及類神經網路理論於偏光膜產品瑕疵檢測與辨識，以電腦視覺檢測系統改善人工檢測之缺失。研究中利用偏光膜影像之影像特性，並結合二元區域樣型（Local Binary Pattern , LBP）紋理特徵，以試誤法找出最佳門檻值，準確地將影像瑕疵區域標記並分割出來。透過十七個有效的特徵值，作為瑕疵分類的依據。瑕疵分類步驟分為三個階段，每一階段皆使用不同的特徵值進行分類，前兩階段主要透過分析類別特性，找出適當的特徵值區分類別，第三階段則使用類神經網路的分類器架構，以七個特徵值作為此分類系統的輸入。本論文所提出的辨識系統可在高類別數、高樣本數的輸入中，產生高辨識率的輸出。實驗結果顯示，本論文提出的偏光膜產品瑕疵檢測與辨識系統平均準確度可達91.55%，可以大大地提升產品檢測之效率。

關鍵字

偏光膜 ； 瑕疵檢測 ； 二元區域樣型 ； 影像分割 ； 類神經網路

並列摘要

In this study, we propose an automatic optical inspection (AOI) system to perform defect inspection for polarizing films. The image processing technologies and artificial neural network are used as the main methodology to improve the disadvantages of human inspection in this system. We use the characteristics in the image of polarizing film with local binary pattern (LBP) texture features to segment the defect region. Seventeen effective features are adopted as the basis for the classification of defects. There are three stages for the defect classification. Different kinds of features are used to classify the defects in each stage. In the first two stages, we look for the suitable features for classification by analyzing the characteristics of different classes. In the third stage, seven features are used as the inputs of neural network for further classification. Experimental result shows that the accuracy of the proposed method is 91.55%. The proposed method also demonstrates the ability of classification with high accuracy for large number of classes and numerous samples. The efficiency of AOI system for polarizing films is significantly improved by our proposed method.

並列關鍵字

Polarizing films ； Defect inspection ； Local binary pattern ； Image segmentation ； Neural network

參考文獻

[22] TFT-LCD產業介紹，林冠宇，高雄縣中山工商。

[24] 黃朝義，偏光膜介紹，光電產業與技術情報，20 期，40-47頁，八十八年。

[2] Xuezhi Yang, Grantham Pang and Nelson Yung, “Fabric Defect Classification Using Wavelet Frames and Minimum Classification Error Training,” Conference Record-IAS Annual Meeting on IEEE Industry Applications Society, May, 2002, Pokfulam Road, Hong Kong, vol. 1, pp. 290-296.

[3] M. C. Hu, I. S. Tsai, “Fabric Inspection Based on Best Wavelet Packet Bases,” Textile Research Journal, vol. 70, no. 8, pp. 662-670, Feb., 2000.

[4] D. M. Tsai, C. Y. Hsieh, “Automated Surface Inspection for Directional Textures,” Image and Vision Computing, vol. 18, no. 1, pp. 49-62, Feb., 1999.