- 學位論文
利用小波轉換為基礎之手機螢幕瑕疵檢測
Using Wavelet Transform for Mobile LCD Module Defect Inspection
若您是本文的作者，可授權文章由華藝線上圖書館中協助推廣。
摘要
本論文的目的在於發展一系統來檢測手機LCD Module 的顯示異常。 隨著科技的進步，手機功能越來越多，在其上的應用也趨向多元，特別是在多媒體的應用上。在手機上LCD Module 可說是一個非常重要的界面，然而目前手機組裝廠用來檢測的方式是利用人的眼睛配合手機畫面的變化來辨別組裝後的LCD Module 是否為正常，不過利用人工來檢查畫面是否正常並不是一種科學的方法，這樣的方法會因為人員的疲勞或是經驗不足而使得檢出率變差更有可能將不良品看成良品；良品看成不良品。所以在手機組裝廠往往必須花費大量的人力來檢查手機。鑑於上述的狀況，本論文透過電腦視覺的各種演算法及技術如紋理分析來發展出一自動偵測手機LCD Module的系統，以簡化手機LCD Module畫面檢測的流程，進而節省人力；而且機器並不會疲勞或是須要有訓練等等狀況出現，所以也不會有檢出力變差或是檢出品質不一的風險。 本論文主要是應用小波封包能量分明的特性來分解紋理影像並找出其能量分佈差異。在本論文中主要是要討論結構性紋理[10]，即透過Wavelet 的分析及配合K-Means 的分群，找出實際上的不良區域。從論文的實驗結果，演算法的平均的Hit Rate 可以達到95.94%，False Alarm Rate 也可以達到0.86%。
並列摘要
The purpose of this thesis paper is to develop a system to inspect the exceptional display of LCD Module of a mobile phone. With advancement of technology, mobile phones have more and more functions and their application is becoming more diverse, particularly in multimedia. LCD Module of mobile is very important component. Currently, inspection of LCD Module is by human eyes and changes of mobile phone pictures. Human inspection on pictures is not a scientific way. Personnel fatigue due to long hours of working or insufficient experiences lead to poor inspection rate or mistake defected products as good products. Therefore, mobile phone assemblers have to use a great amount of manpower to inspect mobile phones. Because of this, based on various algorithms and technologies of computer vision such as texture, this paper develops an automatic mobile phone LCD Module inspection system to simplify the procedures of LCM (LCD Module) pictures and save manpower. Machines do not get fatigued or have insufficient experiences, preventing the risks of poor inspection rate or inconsistent quality. This paper proposes a method to precisely locate defects with mainly wavelet packet energy to decompose grain images and find out the energy distribution differences. The surface texture is divided into two types [10]—statistics texture and structural texture. This paper aims at structural texture to locate the actual defection area through wavelet analysis and K-Means sub-grouping. The average hit rate of algorithm of experiments reaches 95.94% and False Alarm Rate is as high as 0.86%.
參考文獻
[8] 陳錦宗，「小波分析LCD面板瑕疵之研究」，元智大學，碩士論文，民國95年。
[2] Daubechies, I., “Orthonormal bases of compactly supported wavelet,” Communication Pure Appliance Mathematics, Vol. 41, pp. 909-996, 1988.
[3] Mallat, S. G.., “A theory for multi-resolution signal decomposition: the wavelet representation”, IEEE Transactions on Pattern Analysis and Machine Intelligence,Vol. 11, pp. 674-693. 1989.
[4] Bovik, A. C., M. Clark, and W.S. Geisler, “Multi-channel texture analysis usinglocalized spatial filter,” IEEE Transactions on PAMI, Vol. 12, pp 55-73,1990.
[5] Chang T., C. C. Jay and Kuo, “Texture analysis and classification with tree-structured wavelet transform” IEEE Transactions on IP, Vol. 2, pp. 429-441,1993.