特征匹配算法鲁棒性与速度的对比分析

发布时间：2018-03-23 04:00

本文选题：特征匹配　切入点：SIFT　出处：《飞航导弹》2017年01期 　论文类型：期刊论文

【摘要】：特征匹配算法是近年来计算机视觉领域的研究热点。在宽基线匹配、特定目标识别、目标类别识别、图像及视频检索、机器人导航、场景分类、纹理识别和数据挖掘等多个领域,特征匹配算法得到了广泛的应用。为总结特征匹配算法目前存在的问题及指出可能的发展方向,对比分析了当前使用广泛的特征匹配算法的鲁棒性与速度。综述了尺度不变特征匹配算法(SIFT)、快速稳健特征匹配算法(SURF)、二进制稳健尺度不变性特征匹配算法(BRISK)、定向的基于加速段检测子与旋转的二进制稳健独立基本特征匹配算法(ORB)、风式特征匹配算法(KAZE),以及快速风式特征匹配算法(Accelerated-KAZE),分析了算子鲁棒性与速度的影响因素。基于检测子与描述子,利用Mikolajczyk 05标准测试图集与无人机遥感影像测定并分析了复现率、查全率与错误率曲线以及耗时和正确匹配率。实验结果表明,Accelerated-KAZE算法的综合性能较强。
[Abstract]:Feature matching is a hot topic in computer vision field in recent years. In wide baseline matching, specific target recognition, target category recognition, image and video retrieval, robot navigation, scene classification, and so on, Feature matching algorithm has been widely used in many fields, such as texture recognition and data mining. The robustness and speed of the widely used feature matching algorithms are compared and analyzed. The scale-invariant feature matching algorithms such as sift, fast robust feature matching algorithms, binary robust scale-invariant feature matching algorithms and directional invariant feature matching algorithms are reviewed. Based on accelerating segment detector and rotation, binary robust independent basic feature matching algorithm, wind feature matching algorithm, and fast wind feature matching algorithm are proposed. The factors influencing operator robustness and speed are analyzed. Detectors and descriptors, By using Mikolajczyk 05 standard test atlas and UAV remote sensing images, the reproducibility, recall and error rate curves, time consuming and correct matching rate are measured and analyzed. The experimental results show that the performance of Accelerated-KAZE algorithm is better than that of accelerated-KAZE algorithm.
【作者单位】：军械工程学院;
【基金】：国家自然科学基金项目(51307183)
【分类号】：TP391.41

【参考文献】