應用三維傅立葉分析於非同質性隨機紋路表面之瑕疵檢測

Defect detection in inhomogeneous textures using 3D Fourier reconstruction

郭志嘉(Chih-Chia Kuo)

指導教授 ： 蔡篤銘

元智大學/工程學院/工業工程與管理學系/碩士(2002年)

摘要

在工業檢測的領域中，利用機器視覺的自動化瑕疵檢測技術來取代人工檢測，在現今已有很多成功的案例。其原理主要是從待測影像中擷取、描述並分析某些特徵（feature），以自動產生所需訊息。而紋路（texture）則是眾多工業產品之物體表面上的一項重要檢測特徵。傳統紋路分析的應用大多將紋路影像透過傅立葉轉換後的頻譜來擷取與紋路相關的特徵指標，再利用不同分類器，例如貝氏機率、類神經網路等，來進行瑕疵分類或瑕疵檢測，而不同的紋路必須事先設計或選擇個別適用的指標組合。本研究提供一個應用於非同質性隨機性紋路（inhomogeneous texture）表面的瑕疵檢測方法，利用三維影像轉換及三維影像還原的技術來去除數張非同質性隨機影像之間所共有的規律紋路特徵，而不需透過擷取紋路指標的方式來進行瑕疵檢測。
由於非同質性之隨機紋路影像在空間分佈上不具備高度之規律性，本研究將利用序列影像之觀念，將同時取多張正常無瑕疵之紋路影像，於檢測時再將每一張待測影像插入其中而形成一個3D影像資訊，利用正常紋路在不同時間序列上仍應具有類似之規律性的特徵，以3D傅立葉轉換及3D反傅立葉影像還原的技術來偵測非同質性紋路表面上的瑕疵。本研究以觸控面板所使用的濺鍍（sputtering）玻璃基板為實驗樣本，經驗證後發現當序列影像中的不同待測影像彼此之間相似度高時，則本研究方法對於非同質性隨機紋路有良好之檢測效果。

關鍵字

非同質性隨機紋路 ； 表面瑕疵檢測 ； 濺鍍玻璃 ； 傅立葉轉換 ； 影像還原

並列摘要

The purpose of this research aims at the use of machine vision for automatic inspection of defects in inhomogeneous textures such as the sputtered glass substrates of touch panels. The proposed method does not rely on local features of textures. It is based on a global image reconstruction scheme using the 3D Fourier transform. Since a single inhomogeneous texture image does not show homogeneity everywhere in its 2D image, we use a sequence of faultless 2D images to construct a 3D image (the original 2D x-y plane and the additional time-axis) so that the homogeneity of textures can be observed from its time frame.
The method of surface defect inspection is carried out by the 3D Fourier transform of the 3D image and the reconstruction technique of the 3D inverse Fourier transform. By finding an adequate radius in the 3D power spectrum space, and setting the frequency components outside the selected cylinder to zero, we can remove the periodic, repetitive patterns of textures in the time axis. In the reconstructed image, the inhomogeneous region in the original image will have an approximately uniform gray level, and yet the defective region will be distinctly preserved. Experiments on glass substrates in the sputtering process have shown that if each 2D reference image has homogeneity character in its time frame, then our proposed method will have promising results for detecting defects in inhomogeneous textures.

並列關鍵字

Defect detection ； Machine vision ； 3D Fourier transform ； Image reconstruction ； Inhomogeneous textures ； Sputtered substrate ； Surface Inspection

參考文獻

2. 謝志雲，1998年，「應用機器視覺於方向性紋路之表面瑕疵檢測」，碩士論文，私立元智大學工業工程研究所。

4. Amet, A. L., A. Ertuzun and A. Ercil, 1998, “Texture defect detection using subband domain co-occurrence matrices,” Proceedings of the IEEE Southwest Symposium on Image Analysis and Interpretation, pp. 205-210.

6. Bracewell, R., 1956, “Strip integration in radio astronomy,” Aust. J. Phys., Vol. 9, pp. 198—217.

8. Chan, Y. C., K. C. Huang and X. Dai, 2000, “Nondestructive defect detection in multiplayer ceramic capacitors using an improved digital speckle correlation method with wavelet packet noise reduction processing,” IEEE Trans. Packaging and Manufacturing Technology, Vol. 46, pp. 80-87.

9. Chetverikov, D., 1987, “Texture imperfections,” Pattern Recognition Letters, Vol. 6, pp. 45-50.