Current Issue Cover


摘 要
目的 针对直线描述子匹配算法缺乏有效的几何约束,且易受弱纹理、尺度变化的影响,本文提出一种结合多重约束条件的LBD描述子的直线段匹配算法。方法 该算法以LSD算法提取的直线段作为匹配基元,利用SIFT匹配得到的同名点构建同名三角网约束确定候选直线;参考影像上以目标直线段为中心轴建立该直线段的矩形支撑域;根据目标直线段端点及其支撑域四角点在搜索影像上的核线约束建立候选直线段的对应支撑域;利用仿射变换统一目标直线段及候选直线段支撑域的大小;将直线段支撑域分解为大小相等的条形带,通过计算每个条形带的描述符得到该直线段的描述子,依次完成目标直线段与候选直线段LBD描述子的构建;分别计算目标直线段与每个候选直线段描述子向量间的欧式距离,将满足最近邻距离比准则的候选直线段作为匹配结果;最后选取角度约束对匹配结果检核,确定同名直线。结果 实验选取网上公开的3组分别存在角度、旋转、尺度变换的近景影像对作为实验数据,采用本文算法分别对其进行直线段匹配实验,并与其他直线段匹配算法进行对比分析,实验结果表明,本文算法获取同名直线数目约为其他算法的1.06~1.41倍,匹配正确率也提高了2.4~11.6个百分点,从匹配效率上来看,本文算法更为耗时,但兼顾该算法匹配获得同名直线数目、匹配正确率及运行时间,本文算法的鲁棒性更强,匹配结果的准确性与可靠性较高。结论 结合多重约束条件构建的LBD描述子对于存在角度、旋转和尺度变化的影像进行直线匹配过程中具有稳定性。
Based on multiple constraints LBD descriptorand straight line matching

王 竞雪,何 腕营(School;of Geomatics,Liaoning Technical University,Fuxin 123000,China)

Objective A new straight line matching algorithm of the Line Band Descriptor combined multiple constraints is proposed, aiming at the usual problems in many straight line matching algorithms using descriptors, such as the lack of the ultiliazed imformation between matching straight lines, for example, effective geometric constraints and during the matching process, the matching straight line are vulnerable to the influence of the low texture and the scale change of images and so on.Method Straight line segments are extracted from Line Segment Detector method as matching elements, and then corresponding triangulation network which would be established by using SIFT matching points is the constraint region so as to determine candidate lines in the searching image; After the candidate lines selected, the region for band descriptor construction would be built and the construction method would be described as follows: A rectangular support region, which the target straight line segment is the central axis in the region, is established in the reference image; Then the corresponding support region of the candidate straight line segment in the searching image is determined according to epipolar constraints which would be calculated by the endpoints of the target straight line segment and the four corner points of its support region in the reference image; The support regions of the target straight line segment and the candidate straight line segment would be made the same size ultilizing affine transformation; After support regions of straight line segments completed, the regions are divided into a set of bands where each band has the same size and the length of the band naturally equals to the length of the straight line segment, and the Line Band Descriptors of the straight line segment are obtained by calculating the information of each band in support region. There would be a further explanation for the descriptors that it would be calculated by gradient values of four directions of pixels and each band weight coefficient which is along vertical direction in support region would be controlled by gaussian function. Through the above methods, the construction of matching descriptor of the Line Band Descriptor for target straight line segment and candidate straight line segment is completed in sequence; Furthermore, the new Line Band Descriptor combining multiple constraints are normalized to get a unit LBD to reduce the influence of non-linear illumination changes and the descriptor is a forty dimensional vector. The Euclidean distances is the simiarity measure in our algorithm, and the Euclidean distances would be determined by the calculated vectors between the target straight line segment and each candidate straight line segment descriptor. The candidate straight line segment which is satisfied the nearest neighbor distance ratio criterion of Euclidean distances would be the matching straight line as the result. In this process, it is very important to determine the minimum Euclidean distance threshold and the nearest neighbor distance ratio threshold, which directly affect the matching performance of the algorithm. So, to make further improvement on the accurate result, there are many experiments to ensure the multi-threshold. Finally, the angle constraint which is between corresponding straight line and its corresponding epipolar line is used to check the matching result and determine the final corresponding straight lines.Result Three typical groups of close-range image pairs with angle, rotation and scale transformation were selected as experimental data set which is used to complete the straight line segments matching experiments respectively by the proposed algorithm in this paper, and through comparing with other straight line segment matching algorithms, the matching results show that the proposed algorithm in this paper has a strong application in different typical close-range image pairs. Towards experimenting the result analyze,there isconclusion The successfully matches in proposed algorithm are 1.06 to 1.41 times more lines than the comparing other straight line matching algorithms and the proposed algorithm can improve the accuracy of straight line matching by 2.4% to 11.6%. In terms of matching efficiency, although the proposed algorithm is more time-consuming, synthesizing the relevant experiment results of the number of corresponding matching straight lines, matching accuracy and running time, the proposed algorithm are more robust and the straight line matching results are more accurate and reliable. Meanwhile, a high accurate and reliable matching result has been finally obtained.Conclusion The Line Band Descriptor constructed by combining with multiple constraints is stable for line matching of close-ramge images with angles, rotation and scale changes. The instability of other descriptors caused by many factors in line matching is improved.