Multiscaled Texture Synthesis Using Multisized Pixel Neighborhoods

Abstract

Most recent efforts in texture synthesis have focused on using local statistics - that is, selecting and stitching the textures from an input texture sample on the basis of a local color match to enforce textural features either pixel by pixel or patch by patch. A novel texture synthesis method produces high-quality results by introducing a multiscaled texture similarity measurement. Compared with other multiscaled methods, this approach focuses on measuring texture properties at different scales ranging from local to global using an adaptive similarity metric that accounts for texture variations across different image regions