以幾何為基礎的樣本比對

Abstract

樣本比對是一種用來找出在比對的影像中，是否有和樣本(Pattern)相同的實例 (Instance) 出現在影像上。這種方法也可以應用在分析和模型(Model) 有關的物體影像。在我們的方法中，我們使用模型來表示樣本上眾多的取樣點。每一個取樣點都會均勻的分布在邊界上。當比對結束之後，我們會產生一些可能的結果(Result)，這些結果都帶著分數。這些分數代表著樣本和結果之間相似的程度。之後再利用一個適當的門檻值(Threshold)和結果的分數作比較，以便找出真正存在影像上的實例。

Pattern matching is a method for finding the instances of a pattern in matching image. It can analyze an object image relating to a model. In our method, we use a model to represent the pattern which includes many sample points. Each sample point is distributed evenly along the edges. When matching is over, we have a list of results with scores. Each score means the similarity between the pattern and the results. Then, we compare these match scores with an appropriate threshold to decide the results are true or not.