.png){kind=icon}
.png){kind=icon}
Return
Pattern understanding and synthesis based on layout tree descriptor
.png){kind=icon}
PRE
AI
DOI：10.1007/s00371-019-01723-5.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Synthesis from existing examples is a promising way to generate new patterns. However, pattern synthesis is challenging because it is difficult to understand and generate complex structures in patterns. In this paper, we propose an approach based on the layout tree descriptor (LTD) to understand and synthesize patterns from existing ones. The LTD is a binary tree that parametrically describes all primitives, layouts, their dependencies and hierarchies in a pattern. The LTD can be constructed automatically with proposed instance grouping, layout recognition, hyper-primitive matching and tree merging algorithms to realize pattern understanding. To meet specialists' requirements for detailed modification and recombination of patterns, we designed LTD operations including add, remove, replace and grafting operations to allow users to get new patterns by simply adjusting the LTDs. For stylized synthesis, we gave the computing method of LTD similarity. Therefore, the styles of results and input can be compared and users can control generated serialized results by setting the input pattern weights. To meet user's implicit preferences and provide novelty in creative design, we propose an evolutionary approach to creative synthesis. The system generates new patterns continuously based on LTD grafting, meanwhile user selection of preferred patterns will guide the direction of evolution. Experiments using the developed prototype system show that our approach can synthesize novel and complex patterns effectively, meeting different requirements in practice and providing plenty of digital textures for products.
Keywords:
Patterns
Layouts
Synthesis
Design tools
Graphical models
Design space exploration
AI Summary
Key information extracted from the uploaded paper, including a brief overview, abstract, background, key highlights, visual analysis, and future outlook.