自适应变尺度特征点提取方法  

Self-adaptative variable-metric feature point extraction method

作　　者：屈玉福[1] 刘子悦[1] 江云秋[2] 周丹[1] 王一帆[1] 

机构地区：[1]北京航空航天大学仪器科学与光电工程学院,北京100083 [2]宇航智能控制技术国家级重点实验室,北京100854

出　　处：《光学精密工程》2017年第1期188-197,共10页Optics and Precision Engineering

基　　金：国家自然科学基金资助项目(No.51675033;No.51105027)

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。A feature point extraction method for self-adaptative variable-metric constructing image pyramid is proposed to accelerate the feature matching.In this method,number of FAST feature points is adopted as information content quantization in scale space representation and pyramid hierarchy is carried out according to the information difference of blurred images in the neighboring layers.By adjusting scale parameters,Uniform change of detail feature in neighboring images is realized,number threshold of matching points is used to control the height of pyramid and matching efficiency is improved by applying matching instruction strategy named ＂matching and constructing at the same time＂.Last,The contrast experiment is implemented between proposed method and three detection methods-SIFT,FAST,and ASIFT.The experiment results indicate that correct matching rate of the method can reach 43.59％ under various scales.It increase by 25.51％ compared with SIFT.Feature points can still show the targets correctly after they underwent all kinds of changes in lights and angles.The method referred to in the paper selects parameters adaptively according to the feature of target image.It can obtain ideal matching effects without manual adjustmént and adapt to feature extraction and matching in various changeable conditions in high efficieney.

关 键 词：尺度空间 特征提取 特征匹配 高斯图像 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]