AUTOMATIC WANG TILES GENERATION USING PARAMETRIC APPROACH

• Main Author: Astra Pratama, Fairuz
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/68245
• Institution: Institut Teknologi Bandung

In Computer Graphic, Wang-Tile are a method that can be used to tile a texture to a large surface without any obvious repetitive patterns. The main complexity of this method is in the creation of the Wang-Tiles representing a target texture. Current Wang-Tiles creation methods uses non-parametric approaches, essentially randomizing and rearrange a texture to generate a set of Wang-Tiles that looks similar to the original texture and can be tiled on a large surface. These methods can work well on several kinds of texture, but generally have difficulties on structured texture. The usage of parametric approaches has potential to remedy this issue. PSGAN (Periodic-Spatial Generative Adversarial Network) are a parametric model that is trained on a texture, where the resulting model can output similar looking texture with arbitrary size. This model had also been proven to works well on structured texture such as beehives. By adjusting the training process and the random data tensor that’s used by the Generator, PSGAN can be used to create Wang-Tiles from a sample texture. This proposed method can even work on structured or semi- structured textures, outputting Wang-Tiles with better quality compared to existing methods. On the other hand, the model may suffer Partial Mode-Collapse on some textures, and output regular texture tiles instead of Wang-Tiles