摘要
针对近景影像存在的弱纹理、遮挡问题,提出一种基于改进的DAISY描述子和概率松弛的近景影像密集匹配方法。首先,利用SURF提取种子点构建初始视差图,根据影像核线方向改进DAISY描述子的主方向,以影像核线方向的反方向对特征描述子进行掩膜处理,进而对兴趣点进行特征描述。随后,通过松弛迭代的候选点筛选策略渐进地获取正确率占优的特征匹配点。实验表明,相对于传统概率松弛匹配算法,该算法克服了近景影像中弱纹理及遮挡问题导致的误匹配,匹配点数目提高了2倍左右,具有较高的匹配点密集程度和可靠性。
For the problem of weak texture and shade in the close range image,a new dense matching algorithm based on improved DAISY descriptor and probability relaxation is proposed.First of all,the initial parallax figure was built by the seed point extracted using SURF algorithm,the main direction of DAISY descriptor was improved by the image of epipolar direction,and the feature descriptor was masked by image epipolar direction in the opposite direction,in order to obtain the character description point of interest.Then through the relaxation iteration of candidate screening strategy to gradually get dominant characteristics of the correct matching points.Experimental results show that compared with traditional probabilistic relaxation matching algorithm,this algorithm overcomes the weakness in overcoming the blurring image texture and the shade problem caused by false matching,where,the matching point number increases 2times,which has a high degree of dense matching point.
出处
《遥感信息》
CSCD
北大核心
2017年第3期123-127,共5页
Remote Sensing Information
关键词
密集匹配
概率松弛
DAISY
核线倾斜角
候选点筛选策略
dense matching
probabilistic relaxation
DAISY descriptor
epipolar tilt angle
candidate point selection strategy
