摘要
针对传统单一Census变换未充分利用影像信息且精度不高的问题,该文提出了一种结合颜色信息的Census变换半全局立体匹配算法。该算法联合像素点间RGB颜色绝对差值与其Census变换值的匹配代价计算方法,采用八方向半全局视差获取方法获得初始稠密视差图。为进一步提高匹配精度,利用左右一致性交叉检测确定初始视差图中不稳定视差;采用基于均值偏移图像分割的视差优化算法对视差图中不稳定视差进行优化,获得最终视差图。选取4组middlebury立体图片库经典立体像对进行视差获取实验并检测。结果表明,本算法能够获得较高精度且可靠视差图,且在遮挡、视差不连续处的视差精度明显优于单一Census变换立体匹配算法及一些改进半全局立体匹配算法。
Aiming at under utilization of image information and low precision of using Census transformation only in traditional algorithm,a semi-global stereo matching algorithm combining color information and Census transformation was proposed.Joint absolute color difference of pixels in RGB space and Census transformation as matching cost calculation,and initial dense disparity would be obtained by using semi-global algorithm of eight directions.To get a further accuracy of matching,unreliable disparity would be determined by left-right consistency detection in initial disparity.The unreliable disparity would be optimized through the optimizing algorithm based on mean-shift image segmentation and then the final disparity would be obtained.The experiments was implemented through four groups of stereo images form middlebury.The results showed that a high precision and reliability disparity would be obtained and the accuracy of the disparity in occlusion and discontinuity disparity region was high compared with stereo matching algorithm used Census transformation only and some improvement semi-global stereo algorithms.
作者
张晶
王竞雪
ZHANG Jing;WANG Jingxue(School of Geomatics,Liaoning Technical University,Fuxin,Liaoning 123000,China)
出处
《测绘科学》
CSCD
北大核心
2019年第12期147-152,共6页
Science of Surveying and Mapping
基金
国家自然科学基金项目（41871379）
辽宁省教育厅科学研究一般项目（LJYL010）
地球观测与时空信息科学国家测绘地理信息局重点实验室经费资助项目（201801）
