自适应变尺度特征点提取方法

发布时间：2018-03-28 11:24

本文选题：尺度空间　切入点：特征提取　出处：《光学精密工程》2017年01期

【摘要】：为了提高特征点匹配速度,设计了一种自适应变尺度构造图像金字塔的特征点提取方法。该方法采用FAST特征点数量作为尺度空间信息量的度量,利用相邻两层模糊图像的信息量差作为金字塔分层依据,通过调整尺度参数,使相邻图像间的细节特征均匀变化;并使用匹配点数量阈值控制金字塔的高度,设计利用"边匹配,边构造"的图像匹配策略来提高特征匹配的效率。最后,将所设计方法与SIFT、FAST、ASIFT三种特征提取方法进行比较。实验结果表明:所设计方法在变尺度条件下的正确匹配率可以达到43.59%,与SIFT相比提高了25.51%,提取的特征点在目标经历各种光照、角度等变化之后仍能正确表示目标。本文所设计方法根据目标图像特点自适应选择参数,不需要人工调整就可获得理想的匹配效果,能适应各种变化条件下的特征提取和匹配工作,并能提高特征提取和匹配效率。
[Abstract]:In order to improve the matching speed of feature points, a method of feature point extraction based on adaptive variable scale image pyramid is designed, in which the number of FAST feature points is used as the metric of information in scale space. The information difference of adjacent two layers of fuzzy images is used as the basis of pyramid stratification. By adjusting the scale parameters, the detailed features of adjacent images change uniformly, and the height of the pyramid is controlled by the threshold of the number of matching points. An image matching strategy of "edge matching, edge construction" is designed to improve the efficiency of feature matching. Finally, The experimental results show that the correct matching rate of the proposed method can reach 43.59 and 25.51% higher than that of SIFT, and the extracted feature points experience various illumination at the target, the experimental results show that the proposed method can be compared with the three features extraction methods of sift and FASTIFT. The experimental results show that the correct matching rate of the proposed method can reach 43.59 and 25.51% higher than that of SIFT. The method designed in this paper adaptively selects parameters according to the characteristics of the target image, and the ideal matching effect can be obtained without manual adjustment. It can adapt to the work of feature extraction and matching under various changing conditions, and can improve the efficiency of feature extraction and matching.
【作者单位】：北京航空航天大学仪器科学与光电工程学院;宇航智能控制技术国家级重点实验室;
【基金】：国家自然科学基金资助项目(No.51675033;No.51105027)
【分类号】：TP391.41

【相似文献】