CAJ | 학술논문

针对现有频域法 FFT 和 PPFFT 对于大倍率图像配准精度和计算效率难以同时保证的问题，研究了一种基于多重频谱叠加的频域算法，为最大限度减小插值误差噪声，该算法采用多分辨率频谱插值，构造出更接近对数极坐标网格。文中给出分辨率参数设置和算法流程。最后通过算例验证，结果表明：本算法对无裁减图像最大配准倍率可至10，对裁减图像可至5，而FFT和PPFFT（无迭代）对于两种类型图像都约为2；相对于特征点匹配法，本算法也适用于更大倍率的图像配准，从而验证了该算法的准确性和可行性。
침대현유빈역법 FFT 화 PPFFT 대우대배솔도상배준정도화계산효솔난이동시보증적문제，연구료일충기우다중빈보첩가적빈역산법，위최대한도감소삽치오차조성，해산법채용다분변솔빈보삽치，구조출경접근대수겁좌표망격。문중급출분변솔삼수설치화산법류정。최후통과산례험증，결과표명：본산법대무재감도상최대배준배솔가지10，대재감도상가지5，이FFT화PPFFT（무질대）대우량충류형도상도약위2；상대우특정점필배법，본산법야괄용우경대배솔적도상배준，종이험증료해산법적준학성화가행성。
Of frequency-domain method, registration accuracy and efficiency of FFT or PPFFT for large scale image is difficult to be obtained simultaneously. With this problem considered, a new registration method based on overlaying of multiple spectrums is proposed, which interpolates spectrums of multi resolution and generates the grid closest to log-polar, to minimize interpolation error noise. In addition, how to set resolution parameter and the algorithm flow is also explained. Finally, to validate the accuracy and feasibility of the algorithm, three examples were calculated through the new algorithm, and the result shows as followed: as for the image without cutting, the maximum registration scale of the improved method without iterations can reach 10, and as for the image with cutting, the same one can reach 5, while the FFT and PPFFT (non iterative) for the two kinds of image is about 2;compared with the feature point matching method, the improved method is also applied to image with larger scale.