本研究利用街景或測量車等實景影像為參考，改良三維房屋模型的通用紋理貼圖。目的是提高三維數位城市模型視覺化展示時的擬真性，並保有通用紋理貼圖的效能。 研究內容可分成三個部分：(1)影像校正、(2)影像分析與 (3)通用紋理資料庫改良。一開始影像資料需先計算測站、影像與建物三者的共線關係，解算建物在影像之坐標點後視為控制點校正影像幾何。校正完的影像接著可做牆面特徵物分析的演算，解算牆面顏色與窗戶幾何特徵。最後在改良通用紋理資料庫，達到效能友善與部分擬真。 本研究所開發的方法可高度自動化影像前處理、解析牆面窗戶特徵物與牆面顏色，最後再依據這些指標改善原通用紋理資料庫。相較使用隨機挑選的通用紋理，以本研究所開發的方法可達到較擬真的視覺展示，但保有通用紋理對系統的效能友善。;This study tries to improve generic façade textures of building models for realistic look and feel of digital city model visualization. The idea is to use street view or in-situ images as references to produce generic textures that can provide more realistic look and feel of the visualization. The procedure is separated into (1) image pre-processing, (2) image analysis and (3) improving generic texture database. Collinear relationship of stations, images and buildings could help calculate the coordinates, and these coordinates are viewed as ground control points to rectify image geometry. Also, image analysis parses the façade colors and detects window patterns. Finally, using these features, this study improves the generic texture database to maintain efficiency and provide more realistic scene. The algorithm developed in this study demonstrates how to process and parse façade images and to improve generic database systematically. The results demonstrate a better performance and realistic look-and-feel comparing to generic textures chosen randomly.
