Generating facade texture from close-range images for photorealistic city modeling

This paper demonstrates a highly automated image processing system to generate accurate and complete building facade textures from close-range digital images and video data for photo-realistic city modeling. The developed system first automatically or semiautomatically register individual images or video frames with Single-View Metrology or matching of feature points and models. Image stitch is then applied based on matched feature points to form complete facade texture images. Occlusions in the facade texture images are removed and mended with a series of morphological operations or a constrained image inpainting algorithm. The generated facade images can provide realistic and complete facade textures, which enable city models have not only photo-realistic appearance but also accurate texture attributes.