應用小波轉換於紋理瑕疵檢測方法之研究

Wavelet-based Texture Defect Inspection

簡慧文(Hui-Wen Chien)

指導教授 ： 馬杰 ； 陳永盛

元智大學/工程學院/通訊工程學系/碩士(2004年)

摘要

紋理瑕疵即是同中取異，如何判斷是同質，而或是非同質。在此，紋理分析應用即扮演著一個重要的角色。然而，一般常見的瑕疵檢測，多半只是檢測成敗，並未對瑕疵作精確的定位。因此，本論文提出一個從粗略到精細的檢測方法，可以精確檢測出瑕疵位置。主要是應用小波轉換基底分頻解析特性來分解紋理影像並找出其主要存在頻域;並於區塊中套用共變異矩陣理論之統計歸納的有效性，能夠將具有紋理性質的影像歸類出四組代表性的紋理特徵值(ENT, CON, ASM 和 IDM)，藉此作為瑕疵分類的標準依據。最後，採用馬氏矩離分類器，作為區分瑕疵和非瑕疵區域的方法。產品表面的紋理，可分為二類，一是具有「統計性紋理」，二是具有「結構性紋理」。我們將這兩類的影像用來驗証所提出的方法之可行性。傅立葉轉換法和賈伯轉換法在處理紋理瑕疵檢測是一直以來被提出的有效方法。為了驗証我們提出方法的有效性，將以人眼檢測法為標準，來比較分析各方法的實驗結果，加以討論。我們的實驗影像資料庫為40組，在實驗後証明所提出的方法檢測出一個區域瑕疵需耗時449.739毫秒，其準確率為97.286%。

關鍵字

小波轉換 ； 紋理分析 ； 共變異矩陣 ； 馬氏矩離

並列摘要

Defects are detected in homogeneously texture. The way is to find un-homogenous texture. Texture analysis plays an important role in texture defect inspection. In this thesis, we present a coarse-to-fine approach for automatic vision-based inspection system that detects local defect from a textural image. We take advantage of multi-resolution of the wavelet transforms. Furthermore, we divide image into non-overlapping sub-blocks and extract a set of four effective feature vectors (ENT, CON, ASM and IDM) from gray-level co-occurrence matrix (GLCM) of sub-block. Finally, we choose Mashalanobis distance as a classifier to discriminate if the sub-block belongs to defect region. We use statistical textures and structural textures to confirm the feasibility of our approach. Fourier transform and Gabor transform are often used ways to detect defect in textural image. We will discuss and analyze each method. Our experimental images have 40 patterns. The experimental results show our approach to detecting a local defect takes about 449.739ms and its accurate rate is about 97.286%.

並列關鍵字

Wavelet ； Texture ； Co-occurrence matrix ； Mahalanobis Distance

參考文獻

[1] Haralick, R. M., K. Shanmugam, and I. Dinstein, “Textural features for image Classification,” IEEE Transaction on Systems, Man, and Cybernetics, Vol. 3, no. 6, pp.610-621, 1973.

[2] Cohen, F. S., Z. Fan, and S. Attali, “Automated inspection of textile fabrics using textural models”, IEEE Transactions on PAMI, Vol.13, pp. 803—808, 1991.

[3] Daubechies, I., “Orthonormal bases of compactly supported wavelet,” Communication Pure Appliance Mathematics, Vol. 41, pp. 909-996, 1988.

[4] 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.

[5] Bovik, A. C., M. Clark, and W.S. Geisler, “Multi-channel texture analysis using localized spatial filter,” IEEE Transactions on PAMI, Vol. 12, pp 55-73,1990.