摘要
针对传统的Census立体匹配算法对噪声敏感,在视差不连续区域容易出现误匹配的问题,提出了一种基于邻域信息约束与自适应窗口的立体匹配算法.首先,针对传统Census算法对中心像素依赖高的问题,采用邻域十字窗口的加权平均和的方式对中心像素进行赋值.然后,通过设置自适应阈值,将支持窗口的邻域像素与中心像素进行相似性的二次代价计算并与初始代价进行融合,对匹配结果进行进一步约束.在代价聚合阶段,采用颜色阈值不断变化的三约束法进行窗口的构建,并在聚合过程中引入噪声剔除策略.最后,在视差精化阶段采用左右一致性检测与区域投票相结合的方法对视差图进一步优化.使用Middlebury测试平台的标准立体图像进行实验,结果表明:该方法能够有效降低图像对高斯噪声的敏感性,并在误匹配率上低于多种立体匹配算法.
A method based on neighborhood information constraint and self-adaptive window was proposed to solve the problem of mismatching in disparity in discontinuous areas due to the sensitivity of traditional Census stereo matching algorithm to noise.First,in view of the high dependence of traditional Census algorithm of the central pixels,the weighted average sum of neighborhood cross windows was used to assign the central pixel.Second,by setting an adaptive threshold,the secondary cost of similarity between neighborhood pixels supporting windows and central pixels was calculated and fused with the initial cost,and the matching results were further constrained.In the cost aggregation stage,a three-constraint method with changing color thresholds was used to construct windows,and noise elimination strategy was introduced in the aggregation process.Finally,in the disparity refinement stage,the disparity map was further optimized by combining left-right consistency detection and regional voting.The standard stereo image of the Middlebury test platform was used to carry out experiments.Results show that this method can effectively reduce the sensitivity of image to Gauss noise,and the mismatch rate is lower than that of many stereo matching algorithms.
作者
贾克斌
杜奕伯
JIA Kebin;DU Yibo(Faculty of Information Technology,Beijing University of Technology,Beijing 100124,China;Beijing Laboratory of Advanced Information Network,Beijing 100124,China;Beijing Key Laboratory of Computational Intelligence and Intelligent System,Beijing University of Technology,Beijing 100124,China)
出处
《北京工业大学学报》
CAS
CSCD
北大核心
2020年第5期466-475,共10页
Journal of Beijing University of Technology
基金
国家自然科学基金资助项目(61672064)
国家重点研发计划资助项目(2018YFF01010100)
青海省基础研究计划资助项目(2020-ZJ-709)。
关键词
立体匹配
视差图
邻域信息
视差精化
高斯噪声
误匹配率
stereo matching
parallax image
neighborhood information
parallax refinement
Gauss noise
mismatch rate
