應用傅立葉形狀重建技術於物體輪廓瑕疵檢測

Non-referential shape defect detection using Fourier reconstruction

蘇晏徵(Yan-Jheng Su)

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(2006年)

https://doi.org/10.6838/YZU.2006.00040

摘要

在自動檢測應用上，對瑕疵物體的定義往往必須預設一個對應的標準樣本，透過與標準樣本的比較，若待測物體與標準樣本具有相異的特徵表現時，則可將其判定為瑕疵物體，但此應用之前提必須是待測物與標準物完全一致而無變異方能進行比對；而本研究是利用機器視覺技術，對於位置任意旋轉且具形變(deformed)之物件進行外形輪廓瑕疵檢測(shape defect detection)。

本研究主要應用於印刷電路板(Printed Circuit Board)金手指元件的瑕疵檢測。PCB金手指具有形狀不一致(類似但不同)、位置不固定(旋轉)、解析度微小等特性，所以對此類型待測樣本無法挑選特定無瑕疵金手指影像，作為標準金手指樣板(golden template)去判斷個別金手指是否存在瑕疵，因此就每一根單獨支金手指而言無法利用樣板比對(template matching)方法檢測金手指瑕疵，本研究方法主要是根據PCB金手指物體本身的外圍輪廓為基礎，應用傅立葉形狀重建方法，將物件輪廓進行不同程度的還原，藉由還原輪廓與原始輪廓之間的差異，設計差異量測值(特徵值)來判斷此物體是否存在瑕疵。本研究方法發展出四種差異特徵值指標，對五種類型樣本進行測試，其所得到的結果在誤判率(正常視為瑕疵)與漏檢率(瑕疵視為正常)則約小於10%。

關鍵字

印刷電路板 ； 金手指 ； 形狀瑕疵 ； 傅立葉轉換

並列摘要

The traditional technique of pattern matching for shape defect detection requires a reference template to compare with inspection objects. However, the referential approach needs precise alignments of translation and rotation between the reference and the inspection objects. For scene objects with deformed shapes, the inspection task becomes even more difficult, and the traditional pattern matching methods are prone to failure due to lack of standard samples for comparison. This study proposes a non-referential machine vision technique for shape defect detection of objects with rotated and deformed shapes.

The proposed method is implemented for shape defect detection of Printed Circuit Boards (PCBs). Gold fingers on PCBs have the characteristics of deformed shapes, arbitrary orientations, and small sizes. The technique of Fourier shape reconstruction based on the contour of each individual PCB gold finger is applied to detect shape defects. The Fourier reconstruction procedure can restore the contour of each inspection object with varying smoothness. Discrimination features can be extracted from the difference between the original shape and the reconstructed one. The extracted features can then be used as quantitative measurements to evaluate the anomalies on the contour. In this study, four distance measurements are proposed to detect five different shapes of gold fingers. The experimental results from a large amount of test samples showed that the proposed method yielded 90% recognition rates with a single discrimination feature.

並列關鍵字

Machine vision ； Shape defect ； Defect detection ； Fourier transforms ； Printed circuit board ； Gold finger

參考文獻

1.葉繼豪，1997年，「應用共變異數矩陣與小波轉換於BGA基板線路瑕疵檢測」，博士論文，私立元智大學工業工程研究所。

2.林伯聰，2000年，「以資訊理論為基之PCB金屬表面自動瑕疵檢測」，碩士論文，私立元智大學工業工程研究所。

5.Cinque, L., S. Levialdi and A. Malizia, 1998, “Shape desciption using cubic polynomial Bezier curves,” Pattern Recognition Letters, Vol. 19, pp. 821-828.

6.Cinque, L. and L. Lombardi, 1995, “Shape description and recognition by a multiresolution approach,” Image and Vision Computing, Vol. 13, pp. 599-607.

7.Chang, H. H. and H. Yan, 1998, “Vectorization of hand-drawn image using piecewise cubic Bezier curves fitting,” Pattern Recognition, Vol. 31, pp. 1747-1755.