CAJ | 학술논문

针对传统互信息配准方法计算量较大且未利用图像空间信息的缺点，提出了一种结合SURF描述符和广义近邻图的图像配准算法。该算法用SURF从图像中提取尺度空间特征点并获得特征点描述子，然后用广义近邻图来估计Renyi熵与互信息。该算法结合了SURF描述子的鲁棒性和广义近邻图估计Renyi熵的高效性。实验结果表明，对于真实遥感图像，该算法在配准准确度、鲁棒性和速度上都明显优于几种传统配准方法。
침대전통호신식배준방법계산량교대차미이용도상공간신식적결점，제출료일충결합SURF묘술부화엄의근린도적도상배준산법。해산법용SURF종도상중제취척도공간특정점병획득특정점묘술자，연후용엄의근린도래고계Renyi적여호신식。해산법결합료SURF묘술자적로봉성화엄의근린도고계Renyi적적고효성。실험결과표명，대우진실요감도상，해산법재배준준학도、로봉성화속도상도명현우우궤충전통배준방법。
To solve the drawbacks that typical mutual information-based registration has a large amount of calculation neglects the spatial information of images, a new medical image registration method is proposed by combining SURF descriptor and generalized nearest-neighbor graph （GNN）. The algorithm extracts the feature points and SURF descriptor from images firstly, and then uses the generalized nearest-neighbor graph to estimate the Renyi entropy and mutual information. The algorithm combines with the robustness of SURF and the high efficiency of using GNN to estimate the Renyi entropy. The experimental results show that for the real-world remote sensing images, the proposed algorithm can achieve better robustness, higher speed and better accuracy than the traditional methods.