Creating different textile patterns to generate printable fabric images is a difficult image processing task. To accomplish this task, we propose a novel framework for dyeing creation, which allows non-professionals to design individual fabric images. The two main components of this framework are textile pattern discovery and fabric image generation. Since the objects in the fabric image are multi-category and multi-scale, we employ a combination of object pattern and template pattern to discover the repetitive pattern, which can better extract objects and analyze spatial structure. However, the image created with objects and templates cannot be dyed directly, because it does not meet the physical size requirements of dyeing. Therefore, we propose an image super-resolution method for fabric image generation based on edge information prior. It solves the high magnification problem of single image by using deep neural network without training data sets. Extensive experiments on fabric images demonstrate that the proposed algorithm achieves good results both qualitatively and quantitatively. Our method has comparable accuracy compared with state-of-the-art methods and visual results demonstrate our superiority in restoring edges while generating fabric images.

创建不同的纺织品图案以生成可打印的织物图像是一项艰巨的图像处理任务。为了完成此任务，我们提出了一种新颖的染色创造框架，该框架允许非专业人士设计个人的织物图像。该框架的两个主要组成部分是纺织品图案发现和织物图像生成。由于织物图像中的对象是多类别，多尺度的，因此我们采用对象图案和模板图案的组合来发现重复图案，这样可以更好地提取对象并分析空间结构。但是，使用对象和模板创建的图像无法直接染色，因为它不符合染色的物理尺寸要求。因此，我们提出了一种基于边缘信息先验的图像超分辨率方法，用于织物图像的生成。它通过使用深度神经网络而无需训练数据集来解决单个图像的高放大倍率问题。在织物图像上的大量实验表明，该算法在定性和定量上均取得了良好的效果。与最先进的方法相比，我们的方法具有可比的准确性，视觉结果证明了我们在生成织物图像时在还原边缘方面的优越性。