基于语义描述的过程纹理生成

Procedural textures are normally generated from mathematical models and have been widely used in computer games and animations for efficient rendering of natural elements, such as wood, marble, stone and other materials. Although the intuitive way to describe procedural texture is to use semantic attributes, there is no connection between procedural models, model parameters and texture semantic descriptions. In this paper, we propose a novel framework for generating procedural textures according to semantic descriptions. First a vocabulary of semantic attributes is collected for describing procedural textures based on extensive psychophysical experiments. Then a multi-label learning method is employed to label more new textures using the semantic attributes. We construct a procedural texture dataset with semantic attributes and further learn a low-dimensional semantic texture space. Finally, for a set of input semantic descriptions, we are able to find a generation model with proper parameters in this space. This model can be used to generate procedural textures that retain the input semantic attributes. Experimental results show that the proposed framework is effective and the generated procedural textures are correlated with the corresponding input semantic descriptions.