Title: 影像前處理對TFT-LCD面板自動光學檢測瑕疵效力之影響
Examine the effect of the Image Preprocessing on Defect Detection of TFT-LCD Panels by Automatic Optical Inspection
Authors: 黃淑君
Huang, Shu-Chun
Chang, Yung-Chia
Keywords: 自動光學檢測;實驗設計;影像前處理;壞點瑕疵;濾鏡套用;影像尺寸;角度旋轉;Automatic Optical Inspection (AOI);Experimental of Design (DOE);Image Preprocessing;defect pixel;Filter;Image Size;Image Rotation
Issue Date: 2017
Abstract: 利用自動光學檢測技術(automatic optical inspection, AOI) 對薄膜電晶體液晶面板進行瑕疵檢測,可輔助人工檢測的不足,提高面板的檢測效率。AOI利用光學設備取像,再利用演算法判讀影像,進而找出影像的瑕疵;而判讀影像之前對影像進行處理,其目的在於去除雜訊提高檢測效果。本研究針對在應用AOI於檢測液晶面板上的壞點瑕疵時的影像前處理部份進行深入研究,選取影像尺寸、濾鏡套用與角度旋轉三個影像前處理步驟,利用實驗設計法分析其對瑕疵檢出率與瑕疵檢測時間的影響。本研究以國內某電腦品牌公司所提供之真實液晶面板影像為實驗資料,並將壞點瑕疵分為亮點、碎亮點、暗點與碎暗點四類,分別在紅、藍與綠色三種檢測形態下,依照所設計之實驗組合進行實驗與分析。根據實驗結果顯示,濾鏡套用後能顯著的提升瑕疵檢出率且不增加瑕疵檢測時間;對於另紅色碎暗點、綠色亮點與藍色碎亮點瑕疵,影像尺寸大小的改變對瑕疵檢出率有顯著的影響;而角度旋轉則對於瑕疵檢出率與檢測時間無顯著影響。
Applying automatic optical inspection (AOI) on detecting the defects on thin film transistor liquid crystal display (TFT-LCD) panel could compensate the effectiveness of manual panel inspection. AOI uses optical equipment to capture images, and applies specially-designed algorithms to discriminate defects shown. During the AOI process, image pre-processing is an important stage which aims at screening out noises while increasing the detectable rate. This study focused on using design of experiments (DOE) method to examine the effect of image pre-processing techniques on the detectable rate and inspection time for detecting the defective pixels on TFT-LCD panels. The size of input images, the use of filter, and the image rotation are three pre-processing techniques chosen in DOE. The defective pixels are divided into four types: bright, half-bright, dark, and half-dark dots which were inspected under three types of environment: red, blue and green. Real images on TFT-LCD panels provided by a company in Taiwan were analyzed. The result show that the use of filer can significantly increase the defect detectable rate while not increasing the time of inspection. Moreover, for the defects of red half-dark dot, green bright dot and blue half-dark dot, the size of input images has significant effects on the detectable rate. The image rotation, per our experimental results, did not have significant effect on either the detectable rate or the time of inspection.
URI: http://etd.lib.nctu.edu.tw/cdrfb3/record/nctu/#GT070453311
Appears in Collections:Thesis