- July, 2019 : Paper was published in C&G.
- December, 2021 : Web launched.
Textures made of regular repeating patterns are ubiquitous in the real world, most notably in man-made environments. They are defined by the presence of a repeating element, which can show a significant amount of random variations, non-rigid deformations or color noise. We propose an end-to-end pipeline capable of finding the size of the minimal repeating pattern in single images, as well as obtaining the single repetition that, when tiled, produces the most similar synthesis to the complete image. We do this by combining state-of-the-art algorithms in image transformations, repeating pattern detection, image stitching and deep perceptual losses. Additionally, we show how our pipeline can find the minimal color pattern in woven fabrics, which can be useful for both surface-based render methods and computer vision tasks in the textile domain.
Our method can automatically extract the minimum repeatable pattern from a single image, which can be used for texture synthesis:
Using a custom perceptual-based quality metric, we rank potential candidate tiles according to their tileability and how well they represent the input image.
Using a Hough voting space with a multiscale CNN backbone, we can find repeating patterns at multiple abstraction levels.
Elena Garces was partially supported by a Juan de la Cierva - Formacion Fellowship from the Spanish Ministry of Science and Technology.