摘要
尺度不变特征变换(SIFT)图像匹配算法采用高斯差分算子(DoG)进行特征点检测,计算上使用相邻尺度高斯平滑后图像相减。在实践中,检测出的特征点遍布整个图像,造成后续计算量大且误配率高,降低了SIFT算法的实时性。针对以上问题,采用一种优化后的区域检测方法对SIFT特征点检测进行改进。首先利用优化后的区域检测方法检测出目标物体,然后运用DoG算子提取特征点,使特征点集中在目标物体上,从而简化计算,提高SIFT算法的实时性。最后,给出改进算法的实验结果和应用前景。
Difference of Gaussians operator is used in SIFT image matching algorithm to detect feature points. In calculation the adjacent scales Gaussian smoothed image subtraction is used. In practices, the feature points detected are scattered on whole image and that leads to large amount of subsequent calculations and high mismatching rate, arid the real-time property of SIFT algorithm is reduced. To solve the above problems, an optimised region detection method is adopted to improve the feature points detection of SIFT algorithm. First, the target object is detected by an optimised region detection method. Then, the feature points are extracted by Difference of Gaussians operator. It makes the feature points concentrated on the target object, so that simplifies the calculations and improves the real-time property of SIFT algorithm. Finally, the experimental results and application prospects of the improved algorithm are given.
出处
《计算机应用与软件》
CSCD
北大核心
2013年第9期147-150,共4页
Computer Applications and Software
基金
四川省教育厅项目(12ZB070)
绵阳师范学院科研项目(2012A18)
绵阳师范学院自然科学资助项目(2013A12)
关键词
尺度不变特征变换
图像匹配
特征点检测
区域检测
实时性
Scale invariant feature transform（SIFT） Image matching Feature point detection Region detection Real-time property
