基于改进SIFT算法的图像匹配  

Image matching based on improved SIFT algorithm

作　　者：刘佳[1,2] 傅卫平[1] 王雯[1] 李娜[1] 

机构地区：[1]西安理工大学机械与精密仪器工程学院,西安710048 [2]西安科技大学理学院,西安710054

出　　处：《仪器仪表学报》2013年第5期1107-1112,共6页Chinese Journal of Scientific Instrument

基　　金：国家自然科学基金项目(10872160);陕西省自然科学基础研究计划重点项目(2011JZ012)资助

摘　　要：为进一步提高SIFT匹配算法的鲁棒性和正确率,从以下几个方面改进SIFT算法。对图像进行多分辨率小波变换,重建图像近似成分——低频信息参与匹配;采用"回"字形双层方邻窗将特征点邻域区域划分成四部分,建立32维特征点描述符向量;运用欧式距离初步确定匹配点,再用积分图像进一步剔除由于特征点具有空间相似性而出现的误匹配点,从而提高匹配精度。实验表明,本文算法在匹配精度和匹配时间上有明显提高,特别是当图像具有较多局部相似特征时,匹配点数增加,匹配正确率提高。In order to further improve the robustness and accuracy of SIFT matching algorithm, the SIFT algorithm is improved in the following several aspects. Multi-resolution wavelet transform is performed on the images, the image approximation components that are reconstructed-low-frequency information is adopted to match the images;and a ＂ nested hox＂-shaped double square neighborhood window is used to divide the neighborhood of a feature point into four areas and construct a 32 dimension feature descriptor vector. Euclidean distance is used to preliminarily ensure the matching points, and then integral image is used to eliminate the mismatching points caused by the space similari- ty of the feature points, so that the matching accuracy is improved. Experiments show that the proposed algorithm sig- nificantly improves matching accuracy and matching time;especially when the image has more local similar character- istics, all the matching points and the matching correct rate increase.

关 键 词：SIFT算法 小波变换 特征描述符 积分图像 图像匹配 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]