计算机科学与探索 ›› 2017, Vol. 11 ›› Issue (8): 1296-1304.DOI: 10.3778/j.issn.1673-9418.1606018

• 人工智能与模式识别 • 上一篇    下一篇

粗糙集与区域生长的烟雾图像分割算法研究

张  娜,王慧琴+,胡  燕   

  1. 西安建筑科技大学 信息与控制工程学院,西安 710055
  • 出版日期:2017-08-01 发布日期:2017-08-09

Smoke Image Segmentation Algorithm Based on Rough Set and Region Growing

ZHANG Na, WANG Huiqin+, HU Yan   

  1. School of Information and Control Engineering, Xi'an University of Architecture and Technology, Xi'an 710055, China
  • Online:2017-08-01 Published:2017-08-09

摘要: 针对图像型火灾烟雾分割算法不能同时提取白色、灰白色和黑色烟雾的问题,提出了一种粗糙集和区域生长法相结合的烟雾图像分割算法。在RGB颜色空间提取图像的[R]分量,根据R分量的统计直方图构造粗糙度直方图,选取粗糙度直方图中合适的波谷值作为分割阈值,对图像进行粗分割。相对背景图像,烟雾属于运动信息,采用帧间差分法提取运动区域,排除静态干扰。烟雾具有独特的颜色特征,在RGB颜色空间建立烟雾颜色模型,去除颜色相近的运动干扰,获得疑似烟雾区域。在该区域内选择种子点,在粗糙集粗分割的结果上进行区域生长,提取出烟雾区域。实验结果表明,该算法能够同时分割出白色、灰白色和黑色烟雾,烟雾边缘不规则信息保存比较完整,与已有算法的平均分割准确率、召回率以及F值相比,分别提高了19%、21.5%、20%。

关键词: 烟雾图像分割, 粗糙集, 区域生长, 粗糙度直方图

Abstract: In view of the problem that existing image fire smoke segmentation algorithms cannot extract white, gray and black smoke at the same time, this paper proposes a new smoke image segmentation algorithm based on rough set and region growing method. It constructs the roughness histogram according to the statistical histogram of R component extracted from RGB color space of image. And it selects appropriate wave valleys as the segmentation threshold to segment the image in a coarse way. As smoke belongs to motion information compared with background image, the frame difference method is used to extract moving region, excluding static interference. A smoke color model in the RGB color space is established with the unique color characteristics of smoke, to remove the movement interference and obtain the candidate smoke region. The seed points are selected in the region, and region growing method is used to segment image on the results of rough set and extract the smoke region. The results show that the proposed algorithm can segment white, gray and black smoke at the same time, and the edge information of smoke is completed. Compared with the existing algorithms, the mean precision, recall and F-beta of the segmentation are respectively increased by 19%, 21.5%, 20%.

Key words: smoke image segmentation, rough set, region growing, roughness histogram