摘要
为了提高红外图像提升匹配效果,提出了基于人工智能的红外图像特征点匹配方法。利用尺度不变特征变换算法检测红外图像特征点,通过梯度直方图确定特征点方向,完成特征点提取;利用最近邻搜索算法获取特征点的特征向量,求解查询特征向量至原点的欧式距离,搜索查询特征向量的最近邻点,通过设置的搜索范围参数剔除非最近邻点,缩小搜索范围,在该范围内依据查询特征向量至参考向量的夹角,结合匹配规则获取正确最近邻点,即正确匹配向量;通过随机抽样一致性算法去掉误匹配。实验证明:该方法特征点提取的可重复率高于0.5%,具备较优的特征点提取稳定性;可有效匹配红外图像特征点,提升图像清晰度;在不同旋转角度与光照变化倍数时,该方法匹配后的图像灰度分布更均匀,图像信息含量更加丰富。
Research infrared image feature point matching method based on artificial intelligence to improve the matching effect. The feature points of infrared images were detected by scale-invariant feature transformation algorithm, and the direction of feature points was determined by gradient histogram to complete feature points extraction. Using the nearest neighbor search algorithm to obtain the characteristics of the feature point vector, the feature vector to the origin of the query is Euclidean distance, a search query feature vector of the nearest neighbor points, by setting the point of excluding the nearest neighbor search parameters, narrow your search, within the scope of the query based on feature vector to the reference vector angle, combined with the matching rules of getting the nearest neighbor points, That is, the correct matching vector;The random sampling consistency algorithm was used to remove the false matches. Experiments show that the repeatability of feature point extraction is higher than 0.5%, and has better stability of feature point extraction;It can effectively match the feature points of infrared image and improve the image definition;At different rotation angles and illumination change multiples, the gray distribution of the matched image is more uniform and the image information content is more abundant.
作者
韩冰
李印鹏
张丽娟
HAN Bing;LI Yinpeng;ZHANG Lijuan(Hebei Polytechnic Institute,Software Institute,Shijiazhuang 050091,China;Huaxin College of Hebei Geo University,Department of Electronic Information,Xinle City,Shijiazhuang 050700,China)
出处
《激光杂志》
CAS
北大核心
2022年第10期67-71,共5页
Laser Journal
基金
2020年度河北省民办教育协会科研项目(No.HBMXYJJD2009)
河北省教育厅科研项目(No.Z2020229)。
关键词
人工智能
红外图像
特征点匹配
尺度不变
特征变换
直方图
欧式距离
最近邻搜索
artificial intelligence
infrared image
feature point matching
scale invariant
feature transformation
histogram
euclidean distance
nearest neighbor search
