计算机工程与应用 ›› 2020, Vol. 56 ›› Issue (19): 189-196.DOI: 10.3778/j.issn.1002-8331.1908-0239

• 图形图像处理 • 上一篇    下一篇

结合传播滤波的立体匹配算法研究

李婕,巩朋成   

  1. 湖北工业大学 电气与电子工程学院,武汉 430068
  • 出版日期:2020-10-01 发布日期:2020-09-29

Cost Aggregation Based on Propagated Filtering in Stereo Matching

LI Jie, GONG Pengcheng   

  1. School of Electrical and Electronic Engineering, Hubei University of Technology, Wuhan 430068, China
  • Online:2020-10-01 Published:2020-09-29

摘要:

针对局部立体匹配算法在边缘处容易出现误匹配的问题,本文提出了一种结合权值传播进行代价聚合的局部立体匹配方法。首先采用基于颜色梯度的绝对差及Census方法构造了匹配代价函数;然后,引入传播滤波平滑匹配代价的同时保持视差空间图像边缘,与其他局部滤波器相比,该滤波器利用可传播的权值思想,不受传统局部算法窗口大小的影响;最后,通过左右一致性检查和无效视差值填充获得最终视差图。实验表明,该方法在Middlebury Stereo数据集上可获得精确结果,与Middlebury测试平台上的IGF、TSGO和Dog-Guided算法相比平均误差最低。

关键词: 立体匹配, 代价聚合, 传播滤波

Abstract:

To solve the edge mismatching problem in traditional local stereo matching method, a new local stereo matching method is proposed based on weight propagation in cost aggregation step. Firstly, Census combined with absolute difference on color and gradient is employed when computing the matching cost of pixels. Then, propagated filtering is introduced to smooth the matching cost while preserving edges in the disparity space image. The superiorities of this filter are exploits reused weight construction approach for filtering operation comparing with other filters. Finally, several post processing steps are presented to get the final disparity map. Experiments show that the proposed method is able to achieve accurate results on the 30 Middlebury data sets and outperform some other state-of-the-art methods(IGF,TSGO and Dog-Guided) on Middlebury benchmark.

Key words: stereo matching, cost aggregation, propagated filter