使用基於相似度的外觀圖解法於三維物體之形狀記憶與辨識

Abstract

在本論文中，對一個在單純環境下的三維物體，由擷取到辨識出此物體做了一個完整的實現，首先，利用前景偵測結合肯尼邊緣偵測法（Canny edge detection）和加速的梯度向量流動態輪廓偵測法（GVF snake），來得到物體輪廓，接著利用此輪廓找尋出物體的特徵，然後配合計算相似度的方法，代入修改後的Cyr and Kimia的外觀結合演算法（aspect-combination algorithm）和新提出的外觀結合演算法去建資料庫，然後，對於三維物體則依據兩種不同方法所建出的資料庫，利用四種辨識的方法來辨識並比較其結果，最後得出新提出的外觀結合演算法是優於修改後的Cyr and Kimia的外觀結合演算法的結論。

In this thesis, we implement a computational procedure to categorize and classify 3-D objects from their contour under a simple environment. First, we use foreground detection, Canny edge detection and speedy GVF snake to obtain the object’s contour. The contour is then used to establish the object’s features. Two databases are built using methods which compute similarity by modified aspect-combination algorithm proposed by Cyr and Kimia and new proposed aspect-combination algorithm. By using these two databases, we recognize 3-D objects using four recognition methods and compare their performance. Finally, a conclusion is made that our newly proposed algorithm is better than the modified algorithm proposed by Cyr and Kimia.