计算机科学与探索 ›› 2016, Vol. 10 ›› Issue (8): 1154-1165.DOI: 10.3778/j.issn.1673-9418.1508025

• 人工智能与模式识别 • 上一篇    下一篇

纹理图像中重复纹理元素提取方法

杨弄影,李  峰,桂  彦+   

  1. 长沙理工大学 计算机与通信工程学院,长沙 410114
  • 出版日期:2016-08-01 发布日期:2016-08-09

Repeated Texture Elements Extraction from Texture Images

YANG Nongying, LI Feng, GUI Yan+   

  1. School of Computer & Communication Engineering, Changsha University of Science & Technology, Changsha 410114, China
  • Online:2016-08-01 Published:2016-08-09

摘要: 提出了一种交互式纹理图像中纹理元素提取算法,该算法能够在用户提供少量交互的情况下较好地实现纹理图像中重复纹理元素的同时提取。首先采用均值漂移聚类算法将纹理图像分割成独立且连通的子块区域,并构建图像子块区域之间的连通关系;然后结合颜色特征与纹理特征定义一个鲁棒的相似性度量公式,从而准确地捕获具有外观相似特征的纹理元素;在此基础上,通过进一步改进优化的图割模型,最终实现高质量的纹理元素提取。该算法针对前/背景颜色相近的纹理图像中纹理元素的提取有较大改善,并且大大提高了现有图像分割算法的时间效率。

关键词: 重复纹理元素, 纹理特征, 相似性度量, 图割模型

Abstract: This paper presents an interactive element extraction method for texture image, which is able to simultaneously cut out the repeated texture elements with much less user interaction. Firstly, texture image can be segmented into individual and connected subpatches by using Mean Shift algorithm, and their connectivity can be constructed by using Delaunay triangulation. Secondly, by considering color information and texture feature, a robust measurement distance is redefined, which can precisely capture all repeated texture elements with similar appearance. Finally, this paper further modifies the optimization graphcut model in order to achieve the best texture extraction. The proposed method gets a great better extraction result for the texture image that has foreground and background with similar color, and improves the speed of image segmentation significantly.

Key words: repeated texture elements, texture feature, similarity measurement, graphcut model