Procedural Restoration of Texture and Restructuring Geometry From Facade Image

Abstract

We present a novel methodology to procedurally restore facade texture from images, where severe occlusions and lighting variations occur, making the textures difficult to reconstruct effectively. A refinement strategy is designed, which includes an iterative weighted-average algorithm, to restore and update the high-quality consensus texture of the targeted building. Our approach combines component-based structure analysis and rule-based texture recovery techniques that are suitable for removing the occluded areas from a given building facade. We demonstrate our framework on several real-world buildings with varying amounts of occlusion, and show that our approach can be used to generate 3-D buildings with far more aesthetic quality than the previous approaches.