Coherent multi-layer landscape synthesis

Abstract

We present an efficient method for generating coherent multi-layer landscapes. We use a dictionary built from exemplars to synthesize high-resolution fully-featured terrains from input low-resolution elevation data. Our example-based method consists in analyzing real world terrain examples and learning the procedural rules directly from these inputs. We take into account not only the elevation of the terrain, but also additional layers such as the slope, orientation, drainage area, the density and distribution of vegetation, and the soil type. By increasing the variety of terrain exemplars, our method allows the user to synthesize and control different types of landscapes and biomes, such as temperate or rain forests, arid deserts and mountains.

Terrain Super‐resolution through Aerial Imagery and Fully Convolutional Networks

“…We also compared against two recent dictionary‐based terrain amplification methods [GDGP16, AAC*17], although these are concerned with the plausibility of the synthetic results rather than fidelity to ground truth, which had an impact on the quantitative comparison. In [GDGP16], amplification is achieved through a dictionary containing terrain patches at low and high resolution, then finding the representation of the low‐resolution terrain parts on this dictionary and substituting them by the high‐resolution patches.…”

“…These atoms can be stored in a dictionary so that they may be applied to efficient synthesis. By including additional channels and treating them coherently [AAC*17] terrains with multiple properties may be generated, while taking advantage of any information from these additional layers to supplement elevation data.…”

A Review of Digital Terrain Modeling

“…Still, combining primitives to produce large‐scale terrains remains an open research area. Several recent algorithms attempt to amplify large scale terrains with fine‐scale features [GDGP16, AAC∗17]. This is a promising avenue, but it does rely on a reasonable intial coarse terrain to be effective.…”

“…Under this general framework there is, of course, considerable variation between individual methods. For instance, synthesis can involve square patches [ZSTR07, TGM12], individual pixels [GMM15], or circular kernels [AAC∗17], while blending can range from relatively straightforward linear blending [AAC∗17] to graph cuts with Shepard interpolation of the gradient field [TGM12].…”

“…This augmentation of database atoms is extended by Argudo et al . [AAC∗17] to fuse multiple layers, such as base terrain, moisture, soil type and vegetation density. Not only does this allow enriched landscapes, featuring ecosystems and bodies of water, but it also improves the matching to preserve global coherence, such as drainage patterns.…”