By-Example Synthesis of Vector Textures

C. Palazzolo 1 , O. van Kaick 1 & D. Mould 1 1 Carleton University, Canada · 2025

Read By-Example Synthesis of Vector Textures by C. Palazzolo 1 , O. van Kaick 1 & D. Mould 1 1 Carleton University, Canada (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2501-13812.txt.

Introduction

Christopher Palazzolo and co-authors grow new vector textures from a single example picture. Their method takes one raster image and produces a clean, scalable texture of any size, built from crisp shapes rather than pixels. The pipeline starts by cutting the example into its main repeating pieces, called textons, and grouping look-alikes together.

For each group, a descriptor records how the pieces sit beside their neighbors, across all groups. A simple step then tucks the minor pieces behind the main polygons. The background gets a smooth color gradient field, and the minor shapes take colors that blend with it. The authors compare their results with rival methods using a battery of perception-based scores.

Catalog record

CREATORC. Palazzolo 1 , O. van Kaick 1 & D. Mould 1 1 Carleton University, Canada
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE35 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2501-13812

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

By-Example Synthesis of Vector Textures C. Palazzolo 1 , O. van Kaick 1 & D. Mould 1 1 Carleton University, Canada Abstract We propose a new method for synthesizing an arbitrarily sized novel vector texture given a single raster exemplar. Our method first segments the exemplar to extract the primary textons, and then clusters them based on visual similarity. We then compute a descriptor to capture each texton’s neighborhood which contains the inter-category relationships that are used at synthesis time. Next, we use a simple procedure to both extract and place the secondary textons behind the primary polygons. Finally, our method constructs a gradient field for the background which is defined by a set of data points and colors. The color of the secondary polygons are also adjusted to better match the gradient field. To compare our work with other methods, we use a wide range of perceptual-based metrics. \ConferenceSubmission ccs Computing methodologies Texturing † † editors-short: D. Ceylan and T.-M. Li † † editors-star: Y. Gryaditskaya and P. Memari † † editors-tutorial: K. Hildebrandt and R. Mantiuk † † editors-education: R. Kuffner dos Anjos and K. Rodriguez Echavarria † † editors-poster: T. Günther and Z. Montazeri † † editors-dc: L. Liu and G. Singh † † subject: Preprint † † teaser: Left: vector texture synthesis from raster input. Right: post-processing operations facilitated by our vector representation. 1 Introduction Texture synthesis is a long-studied problem, with raster images receiving a great deal of attention. Vector textures, however, are much less explored. One advantage of vector textures is editability [ OBW∗08 ] : numerous existing operations can be used to manipulate [ LTH86 ] and remove elements from a vector texture. Such operations are much more difficult to apply to raster images. Figure shows two examples of synthesized vector textures along with a number of editing operations that are easily applied to polygons, but much harder to apply to a raster image. Previous work has introduced methods for synthesizing simple polygon distributions through polygon packing [ SKA21 , XWL∗23 ] or synthesis of vector patterns from exemplars [ TWZ22 , TWY∗20 , HLT∗09 , AKA13 ] . However, these methods either take simple polygons as input or require the input to be already in vector format with separated vector elements. Other work explores generating vector content that matches a given raster image, but does not permit extrapolating a given exemplar image into a larger texture of a more stochastic nature [ LLGRK20 ] . One might imagine generating vector textures from raster input by first transforming the raster input into vector format with existing tools [ SLWS07 , LLGRK20 ] , then applying the methods discussed above to generate a new vector texture from the input. However, such an approach is likely to fail: vectorization tools often create output that is not separable, with many overlapping polygons exploiting transparency to recreate the input. We propose a novel algorithm for synthesizing vector textures from a given exemplar in raster format (Figure ). Our method is aimed at textures composed of distinct, irregular textons with uneven spacing and lacking global structure, e.g., rust and heterogeneous materials (Figure ). In addition, we assume that the textures lack depth, implying absence of shadows and occlusions. From this input, we generate textures composed of solid-colored polygons on a background gradient field. To synthesize a texture from an input raster exemplar, we build a texton hierarchy through segmentation and extract a background gradient field on which the textons reside. The texton hierarchy can be used to synthesize a novel vector texture of any size. We synthesize a background gradient over a target output domain, and then place textons over the background to replicate a similar texton context as in the exemplar. In addition, we slightly modify polygons to add more diversity to the output. We demonstrate with a set of experiments that our method is able to create textures in vector format that highly resemble the input exemplars, and is competitive when compared to methods that operate directly on the raster domain. Finally, we give examples of how the vector output facilitates editing operations on the textures that are difficult to accomplish with raster images, showing the advantage of the direct synthesis of vector textures from raster images. In summary, this paper makes the following contributions: • We describe a method to convert a natural raster image to a vector representation containing discrete textons and a background gradient. • We propose a method to synthesize novel vector textures given our vector representation. • We demonstrate proof-of-concept editing operations made possible by our discrete texton representation. 2 Related Work To our knowledge, no existing methods directly synthesize a vector texture from a natural raster image. We consider natural textures, i.e., irregular textures with few or no repeated elements. Existing work on vector texture synthesis applies to line drawings and textures that are already in vector format, often with distributions of identical elements. Our method converts a natural image to vector format and then synthesizes a novel vector texture from it. Vector Texture Based Methods . Both Tu et al. \shortcite clusterVectorTexture,ContinuousCurveTextures and Hurtut et al. \shortcite AppearanceGuidedSynthesis use polygons to produce novel textures, but these works assume the example texture is already comprised of polygons (for example, as an SVG file). The objective of these papers is clustering. Tu et al. group visually similar textons, whereas Hurtut et al. build clusters of samples for each vector element using local neighborhoods. Both methods attempt to identify the underlying local distribution of clusters and use this to synthesize novel textures which are visually similar. AlMeraj et al. [ AKA13 ] propose an equivalent to the raster patch-based texture synthesis for vector images, also assuming input primitives are provided in vector form. Saputra et al. [ SKA21 ] and Xue et al. [ XWL∗23 ] describe vector texture synthesis as a polygon packing problem. These methods try to arrange a set of polygons in a finite area while avoiding overlap and reducing empty space. These methods were designed for simple polygon distributions, and may not perform well on a vectorized natural image. Methods also exist for vectorizing raster images. Sun et al. [ SXD∗12 ] describe a scale-independent explicit representation of a diffusion curve image that can be used for mapping textures onto a surface. Li et al. [ LLGRK20 ] propose an algorithm to refine a set of vector graphics primitives to better match a raster image. The output of the method approximates well the input but is composed of overlapping transparent polygons with highly irregular shapes that lack semantic meaning. This makes it difficult to use the polygons for texture generation. Non-parametric Methods . Efros and Leung [ EL99 ] propose a non-parametric sampling approach where pixels are copied from the exemplar directly. Later, Efros and Freeman [ EF01 ] proposed Image Quilting, copying entire patches instead of individual pixels. Such approaches have extremely high quality outputs, but sometimes exhibit visible repetition. Our approach can be considered non-parametric owing to our texton reuse. We provide a quantitative comparison to Image Quilting. Like our work, Galerne et al. [ GGM11 ] and Heitz and Neyret [ HN18 ] propose effective

Read the complete text