摘要
传统SURF算法在对彩色图像匹配时仅基于灰度图像提取单一特征构建特征描述,在特征相似区域易导致误匹配,为提升图像配准精度,提出了一种多特征融合匹配算法。首先利用SURF算法进行特征点检测并构建特征描述符;其次,将彩色图像转换到归一化RG颜色空间,提取特征点邻域内的局部差值纹理信息,叠加到SURF描述符构成改进多特征描述符;计算特征点Hessian矩阵的迹,利用迹的正负对特征点进行分类,并采用最近邻比值法分别对分类的特征点进行对误匹配点进行粗剔除,最后利用改进的RANSAC算法对粗匹配点集进一步优化,计算最佳变换矩阵。实验结果表明,该算法在多种复杂变换下的图像匹配中均能获取较高的匹配精度,与传统算法比较,该算法的平均匹配精度提升了约7%,具有良好的稳定性。
In the process of color image matching,the traditional surf algorithm only extracts a single feature based on the gray image to construct the feature description,which is easy to lead to false matching in the feature similar area.In order to improve the accuracy of image registration,a multi feature fusion matching algorithm is proposed.Firstly,surf algorithm is used to detect feature points and construct feature descriptors;Secondly,the color image is transformed into normalized RG color space,the local difference texture information in the neighborhood of feature points is extracted and superimposed on surf descriptor to form an improved multi feature descriptor;Calculate the trace of the Hessian matrix of the feature points,classify the feature points by using the positive and negative of the trace,and roughly eliminate the wrong matching points by using the nearest neighbor ratio method.Finally,further optimize the rough matching point set by using the improved RANSAC algorithm to calculate the best transformation matrix.Experimental results show that the algorithm can obtain high matching accuracy in image matching under various complex transformations.Compared with the traditional algorithm,the average matching accuracy of the algorithm is improved by about 7%and has good stability.
作者
何显辉
王凯
张平
孙林
HE Xianhui;WANG Kai;ZHANG Ping;SUN Lin(Shandong University of Science and Technology,College of Geodesy and Geomatics,Qingdao 266590,China)
出处
《激光杂志》
CAS
北大核心
2022年第3期87-91,共5页
Laser Journal
基金
国家自然科学基金(No.41771408)
山东省自然科学基金(No.ZR2017MD001)。
