Title:	DESIGN ELEMENTS EXTRACTION BASED ON UNSUPERVISED SEGMENTATION AND COMPACT VECTORIZATION
Author(s):	Hong Qu, Yanghong Zhou, K. P. Chau and P. Y. Mok
ISBN:	978-989-8704-42-9
Editors:	Yingcai Xiao, Ajith Abraham, Guo Chao Peng and Jörg Roth
Year:	2022
Edition:	Single
Keywords:	Image Understanding, Localization, Unsupervised Segmentation, Vectorization
Type:	Full Paper
First Page:	78
Last Page:	84
Language:	English
Cover:
Full Contents:	click to dowload
Paper Abstract:	Repeated design elements are abundant and ubiquitous in decorative patterns, which are now widely used in the process of design for objects, artworks and images found in our living environment. Extraction of repeated design elements from images of existing decorative patterns benefits understanding design, extracting compressed information for subsequent operation, e.g., design generation and vectorization. The early methods based on hand-crafted features were computationally inefficient and less accurate. Most deep learning (DL) based methods focus on natural environment images and are difficult to generalize to decorative images. Besides, DL-based methods require massive datasets with instance-level annotations, which are labor-intensive and hard to get. This paper proposes a novel scheme for design element extraction and vectorization. First of all, unsupervised segmentation is proposed to extract repeated design elements from images of unknown artworks without human assistance. We then distill the color information of the extracted repeated element based on statistics reflected in the color histogram of the input artwork. We develop an algorithm to remove redundant information extracted from images in order to get a compact vectorization result, reusable design element in vector format, at the end. To validate the proposed scheme, we conducted several experiments and the result demonstrated the effectiveness of our scheme and its potential for design generation application.
	Go Back