以密度為基礎之改良型快速分群演算法及其於偏光板瑕疵檢測之應用

Development of an Improved Rapid Density-Based Clustering Algorithm and its Application on Polarizer Defect Inspection

本研究目的為開發即時偏光板瑕疵區域檢測系統，使用線性馬達平台移送偏光板成品，以固定之掃描頻率觸發8k 線型掃描攝影機擷取影像，並且在擷取影像的同時，同步處理部分已經擷取到的影像進行瑕疵區域檢測。在瑕疵區域檢測的流程中，首先使用適應性門檻值進行瑕疵影像分割，再使用本論文所提出的一種以密度為基礎之改良型快速分群演算法 （RDBSCAN），經實驗證明能快速的針對瑕疵影像進行分群處理與位置範圍標記。最後將影像處理與瑕疵標記程序以多執行緒的方式交由雙核心CPU 處理，建立即時檢測之功能。經過實驗結果得知，本研究之雛型系統在檢測平台為70 mm/s的移動速度下，單一CCD 每秒可連續處理寬約57 mm偏光板影像之瑕疵檢測與分群。

This study focuses on developing a real-time polarizer defect inspectionsystem. A polarizer is put on the table driven by a linear motor. An 8k linescan camera is used to scan the polarizer images with constant speed. Thescanned images are synchronously processed for further defect detection.During the image processing, an adaptive thresholding method is used tosegment the scanned images. Subsequently, this study proposes an improvedrapid density-based clustering algorithm (Rapid Density-Based ClusteringApplications with Noise: RDBSCAN) to shorten the time used by usingDensity-Based Clustering Applications with Noise. From the experimentalresults, it can be seen that the proposed method can rapidly cluster thedefective images and correctly mark the defective regions. Finally, the pc-based system employed multi-threading processing technology for imageprocessing and defect marking with a dual core processor to achieve areal-time inspection. From the experimental results, this prototype systemwith single line CCD can continuously inspect a 60 mm wide polarizer at avelocity of 70 mm/s and finish the inspection and clustering.