聚脂膠片表面瑕疵檢測

PET sheet surface defect inspection

王正宇(Cheng-Yu Wang)

指導教授 ： 章明

中原大學/工學院/機械工程研究所/碩士(2016年)

https://doi.org/10.6840/cycu201800035

摘要

產品在製程上經過不斷改良與測試，押出成型後，表面瑕疵為首要判斷產品優劣條件。利用影像瑕疵檢測原理進行產品檢驗，降低目視所造成的誤判。若能使用影像分割法來判別瑕疵的樣貌，就可以在前製程中調整參數，預防問題的發生。
本研究採用影像分割法來檢測表面瑕疵缺陷，利用影像攝影機搭配架設低角度LED光源打光方式取得影像。在分析瑕疵過程中以六道程序來顯現Mura缺陷區域:
①影像的對比度增加。
②再把影像灰階化後添加高斯雜訊。
③利用限制對比適應性直方圖等化方法來改善影像失真的缺點及亮度平衡。
④採用離散餘弦反轉換將頻率域發現的特徵轉成空間域的影像。
⑤並將離散餘弦反轉換與限制對比適應性直方圖等化影像相互相減。
⑥再二值化取得最終所得之影像。
本研究中透過影像分割法，可觀察影像在各項程序呈現的瑕疵狀況，並運用上述所提及的架設方式與六道檢驗程序，以相同參數值對三種素材膠片做瑕疵影像處理。分析後的結果均能將瑕疵顯現且更加呈現瑕疵細微的狀況。以此方式可降低購置設備的成本及目視的誤判，最終達到檢視細微瑕疵分析的成效。

關鍵字

高斯雜訊 ； 限制對比適應性直方圖等化 ； 離散餘弦反轉換 ； 二值化

並列摘要

Products are tested and improved continuously in the manufacturing processes. After extrusion molding, the way to judge the quality of the products is to check if there are any defeats on the surface of the products. Image defect detection methods are applied to check the quality of the product so that misjudgments from visual inspection can be minimized. If image segmentation techniques can be applied to judge the defeats, then we can have parameters adjusted in the pre-process to prevent the misjudgments.
This study applies image segmentation techniques to inspect the surface defeats of the products and uses image camera with low-angle LED light source to get the image. During the process of analyzing defeats, 6 processes are utilized to show the area of mura defeats:
①Increase deposition and evaporation of the image.
②Convert the image to grayscale and then add gaussian noise.
③Apply contrast-limited adaptive histogram equalization method to improve the distortion of the image and brightness.
④Apply inverse discrete cosine transform and transform the image from frequency domain into spatial domain.
⑤Subtract the image of inverse discrete cosine transform and contrast-limited adaptive histogram equalization from each another.
⑥Lastly, apply image binarization to get the final image.
In this study, by using image segmentation technique, the product defeats in each manufacturing process can be observed. Also, we can utilize the techniques and 6 processes mentioned above and use the same parameter to process the image with three different films. After analyzing the results, the defeats can be shown and more details of the defeats can be presented. By doing so, the cost of purchasing new equipment and misjudgment from visual inspection can be reduced. Ultimately, the more details of product surface defeats can be detected.

並列關鍵字

gaussian noise ； contrast-limited adaptive histogram equalization ； inverse discrete cosine transform ； binarization

參考文獻

[9]亞泰金屬工業股份有限公司

[7]李天翔，以影像辨識為基礎之停車場車位管理之研究。大同大學資訊工程研究所碩士論文，2010年。

[2]陳泓翰，液晶顯示器陣列/彩色濾光片製程大面積光阻塗佈狹縫式噴嘴國產化之測試研究。國立交通大學工學院精密與自動化工程學程碩士論文，2010。

[1]Liang-Chia Chen and Chih-Hung Chien and Xuan-Loc Nguyen,“An effective image segmentation method for noisy low-contrast unbalanced background in Mura defects using balanced discrete-cosine-transfer(BDCT),”Precision Engineering,Volume 37,lssue 2,April 2013,Pages 336-334.

[3]游舜豪，應用高解析度線掃描CCD於ITO導電玻璃表面瑕疵檢測之研究。國立台灣科技大學自動化及控制研究所碩士論文，2006年。