Abstract: This paper describes an integrated approach to the construction of textured 3D scene model of buildings from digital images. This approach is implemented in a collection of the global image matching and least Square template matching algorithms. At first, we use improved global image matching algorithm had on variable weight smooth constraint to get the dense disparity surface of the city and the height of every building. Then use a semi-automatic algorithm based on least square template matching to get the accrete edge of buildings. In this approach, the operator recognizes the building and give an approximate location of it in an active mode.The template matching approach automatically searches the exact location of the building's edge. Through this step,we get the vector data of every building in image. At last, we combine both the vector data and texture data in images to reconstruct the textured landscape of the buildings. The potential application include create reality model to be used in city areas and mapping in city area semi-automatically.