We introduce an edge-based procedural texture (EBPT), a procedural model for semi-stochastic texture generation. EBPT quickly generates large textures from a small input image. EBPT focuses on edges as the visually salient features extracted from the input image and organizes into groups with clearly established spatial properties. EBPT allows the users to interactively or automatically design new textures by utilizing the edge groups. The output texture can be significantly larger than the input, and EBPT does not need multiple textures to mimic the input. EBPT-based texture synthesis consists of two major steps, input analysis and texture synthesis. The input analysis stage extracts edges, builds the edge groups, and stores procedural properties. The texture synthesis stage distributes edge groups with affine transformation. This step can be done interactively or automatically using the procedural model. Then, it generates the output using edge group-based seamless image cloning. We demonstrate our method on various semi-stochastic inputs. With just a few input parameters defining the final structure, our method can analyze the input size of \(512\times {512}\) in 0.7 s and synthesize the output texture of \(2048\times {2048}\) pixels in 0.5 s.

我们介绍了一种基于边缘的程序纹理 (EBPT)，这是一种用于半随机纹理生成的程序模型。EBPT 从一个小的输入图像快速生成大的纹理。EBPT 将边缘作为从输入图像中提取的视觉显着特征并将其组织成具有明确确定的空间属性的组。EBPT 允许用户通过利用边缘组以交互方式或自动设计新纹理。输出纹理可以明显大于输入，EBPT 不需要多个纹理来模拟输入。基于EBPT的纹理合成包括两个主要步骤，输入分析和纹理合成。输入分析阶段提取边缘、构建边缘组并存储过程属性。纹理合成阶段通过仿射变换分布边缘组。此步骤可以使用程序模型以交互方式或自动方式完成。然后，它使用基于边缘组的无缝图像克隆生成输出。我们在各种半随机输入上演示了我们的方法。只需几个定义最终结构的输入参数，我们的方法就可以分析\(512\times {512}\)在 0.7 s 内合成\(2048\times {2048}\)像素的输出纹理在 0.5 s 内。