Computer Engineering and Applications ›› 2017, Vol. 53 ›› Issue (24): 213-218.DOI: 10.3778/j.issn.1002-8331.1706-0217

Previous Articles     Next Articles

Method of fast disparity range estimation based on forward search

DONG Benzhi, ZHANG Lijun, JING Weipeng   

  1. College of Information and Computer Engineering, Northeast Forestry University, Harbin 150040, China
  • Online:2017-12-15 Published:2018-01-09

基于前向搜索的快速视差范围估计方法

董本志,张丽君,景维鹏   

  1. 东北林业大学 信息与计算机工程学院,哈尔滨 150040

Abstract: Setting up reasonable range of disparity search can improve the speed and precision in stereo matching. Therefore, this paper proposes an image recursively divided method based on the forward search that estimates the upper and lower limits of the disparity range. Firstly, the reference image is evently divided into a plurality of image blocks, then the current matching block is evently divided with the forward search strategy in the each process of block matching and its sub-blocks are matched with similarity principle. When calculating the upper disparity range, the disparity value of the current block is expressed as the maximum disparity value of the sub-blocks, after finding the maximum disparity image block, it is recursively divided until getting a stable upper limits of the disparity range. When calculating the lower disparity range, the disparity value of the current block is expressed as the minimum disparity value of the sub-blocks, after finding the minimum disparity image block, it is recursively divided until getting a stable lower limits of the disparity range. The experimental results show that the method of using forward search strategy to calculate the disparity range can achieve 28.8% reduction rate of search space while preserving 98% hit rate on average, compared to direct matching, the error rate is reduced by 47.4%, the disparity range is more accurate.

Key words: stereo matching, disparity range, forward searching, recursive subdivision

摘要: 在立体匹配中,设置合理的视差搜索范围能够提高匹配的速度和精度。为此,提出了一种基于前向搜索的图像迭代细分方法用以估算视差范围的上下限。将参考图像均分为若干个图像块,在对每一块的匹配过程中,采用前向搜索策略,对当前匹配块继续均分成若干子块,并对其子块进行相似度匹配。在计算视差范围上限时,用当前块的子块视差中的最大值来表示其视差值,找到视差最大的图像块后继续迭代细分,直到得到稳定的视差范围上限。在计算视差范围下限时,用当前块的子块视差中的最小值来表示其视差值,找到视差最小的图像块后继续迭代细分,直到得到稳定的视差范围下限。实验结果表明,采用前向搜索策略计算视差范围的方法,在搜索空间降低比率达到28.8%的同时能够达到98%的命中率,相较直接进行匹配误匹配率降低了47.4%,能够得到更精确的视差范围。

关键词: 立体匹配, 视差范围, 前向搜索, 迭代细分