Abstract
一般而言，對於工件的辨識方法，在方法上可分為兩大類：統計法則和結構法則。近年有學者提出結合兩者之方法，稱之為屬性字串比對。然而，其方法亦容易受如影像分割等影像前處理問題影響。本研究即針對此缺點，提出一較佳之解決方式，此方法稱之為二階段字串比對法。在第一階段大部比對中，以一適當的特徵做循環字串比對得到輸入物體與參考物體之初步對應關係。接下來即利用此對應關係，做細部比對而求得輸入物體與參考物體之相似函數值。最後，由此相似函數值決定輸入物體所屬種類。除了以二階段字串比對法改進原有方法之缺點外，本研究亦提出支配點偵測方法，解決特徵抽取在影像前處理過程的問題。此外，本研究亦提出一簡單有效的橢圓物體偵測方法。For pattern recognition problems, the two major techniques arestatistical and syntactic (or structural) approaches. Acombined method, namely attributed string matching, has beenproposed by some researchers in the last few years. However,the combined method is also suffered from the segmentationproblem. In this study, we propose a new approach to solve thesegmentation problem. Three areas of mechanical partrecognition are discussed. The first one is a two-stage stringmatching technique. The first stage of matching is to find amatch of vertices between the input object and the referenceobject by using global attributed cyclic string matchingtechnique. In the second stage of matching, a local matching isthen conducted to compute the similarity between two objects.And the input object is thus classified as the reference shapewith maximum similarity. The second area is the dominant pointdetection. In here, we propose a curvature-based polygonalapproximation method to detect the dominant points on curves.The third area is the elliptical object detection problem.