基于改进Census变换和异常值剔除的抗噪立体匹配算法

发布时间：2018-05-31 20:02

  本文选题：机器视觉 + 立体匹配　； 参考：《光学学报》2017年11期

【摘要】：针对Census变换易受噪声影响使得立体匹配算法难以获取高匹配精度的问题,提出了一种改进Census变换和异常值剔除的抗噪立体匹配算法.在初始匹配代价阶段,该方法首先将窗口邻域中值作为参考值并通过映射函数控制异常值,提高了单像素匹配代价的可靠性;然后在代价聚合阶段,对动态聚合窗口中初始代价值进行异常值剔除;最后通过视差计算、视差优化得到最终的视差图.在VS2013软件平台上采用Middlebury标准测试图对初始匹配代价、代价聚合、最终视差图阶段进行测试.实验结果表明,本文算法的抗噪性能优于现有Census变换算法,且错误匹配率达到5.71%.
[Abstract]:To solve the problem that Census transform is easily affected by noise and it is difficult for stereo matching algorithm to obtain high matching accuracy, a new anti-noise stereo matching algorithm is proposed, which is based on improved Census transform and outliers elimination. In the initial matching cost stage, the window neighborhood median is first used as the reference value and the outlier value is controlled by the mapping function, which improves the reliability of the matching cost of a single pixel, and then in the cost aggregation stage, The outlier value is eliminated from the initial generation value in the dynamic aggregation window, and the parallax graph is obtained by the parallax calculation. Middlebury standard test diagram is used on VS2013 software platform to test initial matching cost, cost aggregation and final parallax phase. Experimental results show that the proposed algorithm has better anti-noise performance than the existing Census transform algorithm, and the error matching rate is 5.71%.
【作者单位】： 上海大学通信与信息工程学院;
【基金】：国家自然科学基金(61471230)
【分类号】：TP391.41

【相似文献】

中国期刊全文数据库 前10条

1 雷磊;郑江滨;宋雪梅;;基于改进Census变换的立体匹配算法[J];计算机应用研究;2013年10期

2 王军政;朱华健;李静;;一种基于Census变换的可变权值立体匹配算法[J];北京理工大学学报;2013年07期

3 宁晓斐;胡波;赵磊;_澄闹，

本文编号：1961108