摘要
当前多数局部不变特征算法主要针对刚性形变的图像匹配问题,但非刚性形变普遍存在且多数表现为各向互异。提出一种面向各向异性非刚性形变的局部不变特征算法。本算法对最稳定极值区域进行同性化处理,使其各向形变在一定程度上趋于同性且不受方向性影响,然后将图像投影到三维空间中,并通过调整外观权值获得形变不变的测地距离,最后建立各向异性测地距离灰度直方图(AGIH)描述符,使本算法进一步适应各向异性的微小形变。验证实验表明,在非刚性形变情况下,与SIFT算法、GIH算法相比,本算法在相同的错误匹配率的情况下,正确检测率平均提高了10.65%。
Recently, the research work on local invariant features mainly focuses on obtaining the invariance to rigid deformation. However, the intrinsic appearance of an object under various view conditions is of the non-rigid deformation, especially the anisotropic one. Therefore, we propose a method using anisotropic non-rigid deformation invariant features. First, we conduct an isotropic deformation on the Maximally Stable Extremal Region, and rotate the corresponding pairs to the same orientation. Then, the original image is projected into 3D space. With the increase of the aspect weight, numerous sets of deformation invariant geodesic distance are obtained. At last, the Anisotropic Geodesic-Intensity Histogram is adopted as a non-rigid deformation invariant local descriptor to make the proposed algorithm fit anisotropic small deformation. Evaluations have been conducted on the non-rigid deformation dataset against Scale Invariant Feature Transform and Geodesic-Intensity Histogram. Result shows that the correct detection rate can be improved by about 10.65% under the same false positive rate.
出处
《仪器仪表学报》
EI
CAS
CSCD
北大核心
2010年第1期99-104,共6页
Chinese Journal of Scientific Instrument
基金
国家863高技术研究发展计划项目(2007AA01Z423)
国防"十一五"基础科研基金项目(C10020060355)
国防基础科研基金项目(CS-OC2)
重庆市科技攻关计划项目(CSTC
2007AC2018)资助
关键词
局部不变特征
非刚性形变
各向同性
各向异性
AGIH描述符
local invariant feature
non-rigid deformation
isotropic
anisotropic
anisotropic geodesic-intensity histogram
