匹配面圖元進行相對方位自動化之研究

A Study of Matching Area Features for Automatic Relative Orientation

湯凱佩(Kae-Pey Tang) ； 曾義星(Yi-Hsing Tseng)

《航測及遙測學刊》 2卷2期 (1997/06) Pp. 1-13

https://doi.org/10.6574/JPRS.1997.2(2).1

摘要

本文提出一個匹配立體像對間之面圖元的方法來自動恢復相對方位的方法。首先，將影像以影像金字塔（Image Pyramid）的方式儲存，在最粗層（低解析度）影像中以區塊成長法萃取出均調區塊之面圖元，並以Fourier描述元描述其邊界線，而後利用最小二乘匹配Fourier描述元的理論，計算所有可能對應面圖元之形狀相似性及方位關係來作為辨識資訊，根據辨識資訊，應用Hopfield-Tank類神經網路進行面圖元匹配，進而建立影像間近似的平面轉換關係，並以此來預估影像間近似的共軛位置，協助區域匹配法求取共軛點位。所得的點位將由粗至精逐層匹配求出精確的共軛點位以完相對方位的建立。最後，本文以一空照立體像對進行實驗測試，其成果顯示此方法完成全自動化相對方位的可行性。

關鍵字

圖元匹配 ； 自動相對方位 ； 影像匹配 ； Fourier描述元 ； 類神經網路

並列摘要

This paper presents a fully automatic method to reconstruct the relative orientation of a pair of stereo images by matching area features. First, by using the concept of coarse-to-fine process, the stereo images are stored in a form of image pyramid. In the coarsest level, area features of homogeneous regions are extracted by using region-growing method. Then, their boundaries are described with Fourier descriptors. Applying the least-squares approach to matching Fourier descriptors, the recognition information consists of shape similarity and relative orientation between each possible matching pairs of area features is obtained. Based on the recognition information, conjugate area features can b e matched by using Hopfield-Tank neural networks. When conjugate area features are determined, the approximate orientation between images can be easily solved. Knowing the approximate orientation, area-based matching, through a coarse-to-fine matching process, can be applied to obtain accurate tie points, and relative orientation can be solved. A pair of real aerial images were tested. The results show a successful application of the method.

並列關鍵字

Feature Matching ； Automatic Relative Orientation ； Image Matching ； Fourier Descriptiors ； Neural Networks