一個簡單快速及精確之影像接圖方法

A Simple, Fast, and Accurate Method for Image Registration

李超良(Chau-Liang Lee)

指導教授 ： 林啟芳

元智大學/工程學院/電機與資訊工程研究所/碩士(1995年)

摘要

在本篇論文中，我們提出一個連接兩張部份重疊影像的方法。所提
出的方法適用於連接兩張不同方向、位移和微量放大縮小的影像。主要步
驟包含下列兩項：(一) 抽取特徵，及 (二) 建立對應關係。在第一個步
驟中，我們設計一個精密的圓形區域來抽取特徵。此圓形區域從左至右及
從上而下掃描這些影像，並將圓心區域內灰階值最高的點選為特徵點，在
此步驟完成後，將會獲得到兩個特徵點集合，分別對於此兩張影像。
在第二個步驟中，這兩個點集合之間的對應問題將被解決。我們利用在
[12]中所提出的鬆馳法，並提新的改善方法以增進鬆馳法的效率。我們也
發展出一個新的方法(最小損失選取法; 簡稱為MLS)來決定"最佳"點配對
，最後利用[12]中的方法來計算出幾何轉換係數。實驗結果可顯示所提方
法的優越性。

關鍵字

影像接圖 ； 特徵抽取 ； 點配對 ； 鬆馳法

並列摘要

In this study, a method to register two partially
overlappingimages is presented, which is applicable to the
registration ofimages with variant orientations, translations,
and small amountof scale changes. The method is designed to
consist of thefollowing two steps: (1) to extract features, and
(2) to establishcorrespondence. In the first step, a weighted
circular window isprecisely designed to extract features, and
the window is movedpixel-by-pixel from left-to-right and top-to-
bottom. The pixel atthe window center is treated as a feature
point if its intensitysatisfies the constraint of local maximum.
When the registering areas of the two images are scanned
completely, two sets of pointsare obtained accordingly, and the
process of the first step is completed. In the second
step, the correspondence between the two pointsets are
determined. The problem is tackled by utilizing the relaxation
method presented in [12], and an improvement method isdeveloped
in this study to improve the performance of the
relaxationscheme. A heuristic method named MLS (Minimum Loss
Selection) methodis also developed by us to decide the *best*
matching pairs betweenthe two sets of points, and the
transformation parameters are derived utilizing the method
presented in [12]. The performance ofthe method is demonstrated
by experimental results.

並列關鍵字

Image Registration ； Feature Selection ； Point Pattern Match ； Relaxation Scheme