改进的MOPs图像匹配算法

Image Matching Algorithm Based on Rotation Invariant Descriptors

针对MOPs算法在处理图像旋转方面的不足,提出了一种新的旋转不变的特征描述子,用以匹配存在缩放、旋转、平移的图像。用Harris角点检测算法,检测金字塔模型的每层图像,在特征点处提取旋转不变特征描述子,用特征描述子之间的匹配来寻找两图像间的匹配点,最后用RANSAC方法剔除误配点,实验结果表明比MOPs算法具有明显的优越性。

In order to solve the problem of image rotation that existing in MOPs algorithm, a novel rotation invariant descriptors are presented to match images with transform of zoom, rotation, translation. Use Harris comer algorithm to locate feature points in pyramid modal, and then rotation invariant descriptors are extracted, match the descriptors to obtain initial match points, inlier matches are removed using RANSAC algorithm and affine parameters are calculated. Experimental results show that our algorithm is better than MOPs.