對一個三維的立體物而言,在不同的角度可以得到不一樣的物體影像。在本論文中,我們將探討以MACE理念所設計出的濾波器,並將之應用在3D旋轉的目標識別上。本研究的重點是設計一個濾波器,使其在對三維旋轉的物體影像辨別能力上能得到一最佳的識別效果。 首先,我們提出一個以NOJTC為架構的方式,來做光學的影像識別。此架構具有位移的不變性,且對影像的扭曲有不錯的容忍度,因此成了將此架構用在影像識別上的有利因素。在許多先前已被發表的論文中,可看到以此NOJTC架構並結合MACE function所設計出的濾波器,在針對單一平面旋轉影像上的確有不錯的表現。所以我們以此為基礎,加以延伸到對物體三度空間的旋轉影像識別上。經由數值計算的結果,我們可以看到在不同設計條件的情況下,得到的不同辨別表現。最後,我們也對原始影像加入了雜訊和背景,來看看其在辨別效果上的差異。
For a 3D image, there are different views with various angles. In this thesis, we present a design of minimum average correlation energy (MACE) filter function to recognize a target with 3D rotational views. Our research is emphasizing on optimum discriminant capability. First, we propose a method that uses nonzero order joint transform correlation (NOJTC) along with Lagrange multipliers technique for image recognition. The NOJTC has a powerful capability to achieve shift-invariance and a potential for distortion-tolerant image recognition. It has been shown that relative high correlation peak can be observed. Then, we extend this concept to design JTCs to perform 3D rotational-invariance. From numerical results, we can see that discriminant performance. Finally, we add noise and background to see the results of recognition.