UML線上手寫辨識系統

Abstract

在本論文中，我們建構了一個UML 線上手寫辨識系統。 根據我們的觀察，UML 的圖形多半為類似方形或是菱形的圖 形，因此在本系統中利用決策樹的方式，來達到辨識的效 果。首先我們擷取使用者輸入圖形的幾何特徵，來進行第一 階段的分類。接著從輸入圖形擷取我們需要的特徵，和其所 屬分類中各個圖形的特徵向量進行比對，即可得到最後的辨 識結果。本系統之優點在於可以接受使用者任意筆順的輸 入，並且辨識的方法較之前更為簡單有效，正確結果出現在 前三名的辨識率為91.24%。

In this thesis, we construct an online handwritten recognition system of UML diagrams. We use a decision tree to do recognition. According to our observation, the shapes of the notations of UML diagrams almost look like rectangles or diamonds. Based on this characteristic, an input notation is first classified to the correct category. Then some notation features are extracted from the input notation and used to do final recognition. The advantages of our system are that we can accept free style input and our method is simpler and more efficient than previous methods. The recognition rate of the top three choices is 91.24%.