电子学报 ›› 2016, Vol. 44 ›› Issue (7): 1649-1655.DOI: 10.3969/j.issn.0372-2112.2016.07.019

• 学术论文 • 上一篇    下一篇

基于集成学习的交互式图像分割

刘金平1, 陈青2, 张进2, 唐朝晖2   

  1. 1. 湖南师范大学数学与计算机科学学院, 湖南长沙 410081;
    2. 中南大学信息科学与工程学院, 湖南长沙 410083
  • 收稿日期:2015-01-27 修回日期:2015-05-12 出版日期:2016-07-25 发布日期:2016-07-25
  • 作者简介:刘金平 男,1983年生于湖南邵阳.湖南师范大学数学与计算机科学学院讲师.研究方向为计算机视觉与模式识别.E-mail:ljp202518@163.com;陈青 女,1967年生于湖南长沙,中南大学信息科学与工程学院博士研究生,研究方向为计算机信息技术、智能自动化信息检测与处理等领域的研究工作
  • 基金资助:

    国家自然科学基金(No.61472134,No.61171192,No.61272337);湖南师范大学青年基金(No.11405)

Interactive Image Segmentation Based on Ensemble Learning

LIU Jin-ping1, CHEN Qing2, ZHANG Jin2, TANG Zhao-hui2   

  1. 1. College of Mathematics and Computer Science, Hunan Normal University, Changsha, Hunan 410081, China;
    2. School of Information Science and Engineering, Central South University, Changsha, Hunan 410083, China
  • Received:2015-01-27 Revised:2015-05-12 Online:2016-07-25 Published:2016-07-25

摘要:

针对交互式图像分割人工标记示例匮乏、不同目标区域难以均衡标记,单一分类器难以获得有效分割结果的问题,提出一种多分类器集成学习的交互式图像分割方法.采用多元自适应回归样条(MARS)方法构造第一个分类器;同时引入光滑薄板样条回归函数(TPSR)构造与之互补的第二个分类器,综合组成bagging集成学习器,以降低单一分类器对噪声的敏感度并进一步提高人工标记样本特征空间的利用率.随后,基于半监督学习中的聚类假设,结合bagging多学习器并联特点,提出一种REG-Boosting半监督学习算法,实现半监督图像分割.在不同数据集上的验证性和对比性实验表明所提方法的有效性和优越性.

关键词: 交互式图像分割, 多元自适应回归样条, 集成学习, 薄板样条回归, 半监督学习

Abstract:

A kind of interactive image segmentation method based on ensemble multi-classifiers is put forward to solve the problem of unsatisfactory segmentation results based on scarce or unbalanced labelling labels on different object areas by single learner.The first classifier is established based on multivariate adaptive regression splines (MARS) method.A complementary thin plate spline regression (TPSR) classifier is simultaneously established.By combination of these two classifiers,a bagging ensemble learner is achieved to reduce the noise sensitivity and make further efforts of improving the use of the feature space information of the labeling samples.Ultimately,a kind of REG-Boosting algorithm for semi-supervised image segmentation is put forward based on the clustering hypothesis in the ensemble learning combining with the parallel characteristic of the bagging multi-learners.Abundant validation experiments and comparative experiments on different test sets confirm the effectiveness and out-performance of the proposed method.

Key words: interactive image segmentation, multivariate adaptive regression splines(MARS), ensemble learning, thin-plate spline regression (TPSR), semi-supervised learning

中图分类号: