Improvement of image registration algorithm is beneficial to enhance image precision in various application fields. By treating medical CT image of the human hands as the research object, hybrid attributes composing of pixel gray value, neighborhood mean value and normalized neighborhood variance are innovatively applied into extraction are of image contour. Also, a concept of principal axes is adopted to calculate the centroid and the angle of axes and axis between reference image and floating image, for realization of image registration through obtained registration parameters. Research shows that the adhibition of multi-attribute feature is helpful for further improvement of precision of contour extraction; compared with the traditional Canny method, the error of image registration algorithm based on principal axes is reduced by 70% or more, indicating the feasibility of the proposed algorithm, and the high promoting value in fields including medical image.%图像配准算法改进有利于提高各大应用领域的图像精度。以人手部CT医学图像为研究对象，首次在像素灰度值的基础上融合了邻域均值和归一化邻域方差属性，实现了图像轮廓提取；并应用力矩主轴思想，计算了参考图像和浮动图像的质心、主轴与坐标轴的夹角；通过所得配准参数实现了图像配准。研究表明：多属性特征的引入有助于进一步改善轮廓提取精度；相比传统的Canny法，基于力矩主轴的图像配准算法误差降低了70%以上，表明了算法具有可行性。值得在医学图像等领域加以推广。
展开▼
