朱程涛, 李锵
AuthorsHTML:朱程涛, 李锵
AuthorsListE:Zhu Chengtao, Li Qiang
AuthorsHTMLE:Zhu Chengtao, Li Qiang
Unit:天津大学微电子学院，天津 300072
Unit_EngLish:School of Microelectronics, Tianjin University, Tianjin 300072, China
Abstract_Chinese:在立体匹配算法中, 视差精炼过程是影响立体匹配精度的重要步骤．针对传统基于权重中值滤波的视差精炼算法, 其滤波性能不佳导致精炼后的立体匹配精度不高的问题, 本文提出了一种基于梯度域引导滤波的视差精炼迭代算法．该算法首先依据梯度域引导滤波的原理, 将其应用于视差精炼的过程, 得到改进后的权重中值滤波算法; 同时对视差图进行迭代精炼, 以提高立体匹配精度．实验仿真结果表明, 相较于传统方法, 本文算法能够有效地降低视差精炼的平均误差, 获得较高精度的视差图．
Abstract_English:Disparity refinement is a critical step for high-quality depth map in stereo matching. Recent approaches using weighted median filters，such as bilateral filter and guided image filter，remove streak-line artifacts with limited performance. An effective disparity refinement algorithm is proposed in this paper，which is based on the technique of gradient domain guided image filtering and further improves the refinement performance. The algorithm is a combination of the gradient domain guided image filtering and the weighted median filtering that can be applied iteratively to improve the accuracy of refinement. Experiments on the Middlebury benchmark demonstrate the superior performance of our approach.
Keyword_Chinese:立体匹配; 视差精炼; 梯度; 图像滤波
Keywords_English:stereo matching; disparity refinement; gradient; image filtering

PDF全文下载地址:http://xbzrb.tju.edu.cn/#/digest?ArticleID=6043