以色彩資訊協助影像特徵之萃取

Features Extraction from Color Information

蔡玉慧(Yu-Hui Tsai) ； 王蜀嘉(Shue-Chia Wang)

《航測及遙測學刊》 4卷2期 (1999/06) Pp. 1-14

https://doi.org/10.6574/JPRS.1999.4(2).1

摘要

欲於大比例尺航空像片中自動辨識出房屋並重建其三維的幾何模型時，第一步的工作就是要先找出何處可能有房屋。有很多方法都可用來近似找出可能是房屋的地點，但是當航空影像為彩色時，則顯然還可用彩色資訊來協助找出房屋之位置。因為在一般的情況下單獨的一個屋頂大多具有一致的顏色，且與周圍的背景，如草地、柏油路面的顏色有很大的差異。本研究由兩個方向來研究色彩資訊所可能提供的輔助。第一個方向是應用色彩資訊增加線特徵萃取的完整性，另一個研究方向則是以色彩資訊直接協助房屋面狀區塊之萃取。為了測試本文所提方法之可行性，我們選用了三種類型的房屋影像來做實驗，其型態為獨立房屋、連棟屋、以及高樓。由實驗結果得知，色彩資訊確實可增加線特徵萃取之完整性。而在面狀建物區塊的萃取方面，當屋頂之色彩一致時，可成功地找出建物之位置及近似邊界線。證實了色彩確可輔助影像特徵之萃取。

關鍵字

彩色資訊 ； 特徵萃取 ； 色彩模型 ； 區塊成長 ； 影像分類

並列摘要

The first of the automatic reconstruction of buildings from aerial images is to find where the buildings are. Many methods can be used for this work, but if the aerial photo is color, it is a very useful information for finding the houses because usually the color of one single roof is uniform and is different from the color of adjacent objects likes grass, roads, etc.Two aspects are researched here to study the possible usage of color information. First, we try to use color information to enhance linear feature extraction. Second, we try to omit the step of linear feature extraction and use the color information to extract building regions directly. In order to test the feasibility of the thought, we choose three different types of building images for experiment. The first type includes isolated buildings, the second type includes aggregation of buildings, and in the third type there are high apartment buildings. The results of the experiment show that in one hand color information do help to extract more complete linear features, and in the other hand, color information can very effectively be used to extract building regions, presumed the color of each individual roof is uniform. It proves that color information is indeed helpful for feature extraction.

並列關鍵字

Color information ； Feature extraction ； Color model ； Region growing ； Image classification