中文摘要：

为了解决喷墨印花纹理图像在噪声环境下配准精确度低，以及局部区域存在大的配准偏差的问题，提出一种新的基于单元分解光流场的图像配准算法．利用单元分解技术对光流场图像配准问题进行建模，采用阶谱分层策略和基函数自适应调整，对单元分解过程中生成的局部和全局网格实施灵活度控制．通过引入一个新的能够对纹理结构的光滑效果实施约束的特征能量项，可以使网格单元取得令人满意的纹理表征效果，并且能够提高光流场模型在定位精细纹理边缘时的配准精度．对含噪喷墨印花纹理图像的配准实验结果表明了本文算法的可行性．

英文摘要：

A novel image registration algorithm based on the unit partitioning of optical flow field was proposed in order to solve the problems of the low accuracy in digital registration of ink-jet printing images under noisy environment and the large registration deviation for local areas. The technology of unit partitioning was applied to modeling the problem of optical field image registration. Hierarchical strategy and adaptive adjustment of basic function were proposed to implement the flexibility control on the local and global meshes generated by the processing of unit partitioning. A novel feature energy term restricting the effective of smoothing on texture structure was introduced in order to obtain satisfied efficiency of texture characteristic for mesh elements and improve the registration accuracy of optical flow model for positioning fine texture edges. Experiments on noisy ink-jet printing texture images were presented to illustrate the feasibility of the algorithm.