Deep Terrain Expansion: Terrain Texture Synthesis with Deep Learning

Abstract

In real-world applications terrains play a cardinal role in the field of games and geospatial applications such as Geographic Information Systems (GIS). The textures of a terrain are essential for creating virtual photorealistic environments for users. In many cases, the entire texture of a region is not available in high resolution or is much smaller than the required texture to cover a terrain. Tiling of a texture across a terrain or using an enlarged version of it usually fails to provide an acceptable photorealistic result. Consequently, high quality texture synthesis is a central issue in such settings. In this paper, we explore a novel methodology that extends previous work providing both synthesis and expansion/shrinkage of a texture.