This study developed a highly automated video processing system for photorealistic visualization of three-dimensional (3D) building models from close-range video sequences. The objective was to produce complete and seamless photo-realistic building facade images and to map them onto correct building model facets effectively and efficiently. Emphases were placed on two important tasks. The first was to merge overlapped video frames to generate mosaicked texture images that were continuous in both geometric outlines and in color shadings. The second was to remove unwanted texture regions, including shadows and areas blocked by road trees and other foreign objects, and to reconstruct the removed texture blocks from near by texture information. The developed system first identified interest points on overlapped frames extracted from video sequences. The interest points were then filtered and matched using a developed algorithm based on the Normalization Cross Correction operation to generate seamless texture mosaics. Then, a combination of morphological image processing and other techniques were applied to the generated texture images to automatically identify and remove unwanted texture regions. Video morphology was also used to reconstruct the removed texture blocks by mirroring nearby texture information to produce complete facade textures. Finally, the texture mosaics were mapped onto corresponding building model facets with parametric linear transformation to for a photo-realistic visualization of 3D building models.