一种引入傅立叶-梅林-仿射两级变换的影像区域匹配框架

针对影像区域匹配方法几何形变敏感的应用局限性,将特征匹配、相位匹配的基本思想引入区域匹配过程,基于傅立叶-梅林-仿射两级变换建立了归一化互相关灰度相似性计算下的自适应模板匹配框架,并详细阐述了该框架下的全局运动估计、模板“粗”纠正与搜索预测、局部仿射变换下的“精”模板动态生成等关键过程与算法。实验证明了该方法的有效性。

One New Area-based Stereo Matching Method Based on Fourier-Mellin-Affine Two-stage Transformation

An adaptive template-matching framework integrating the Fourier-Mellin and Affine transformation is proposed for implementing area-based matching toward complicated image motion or geometry distortion.Firstly,IR(interest of region) with remarkable intensity distribution is automatically generated from reference image by dividing image into grids and within each grid,applying the Moravec operator to extract the most distinct point as the center of IR.Secondly,based on Fourier-Mellin transformation,global image motion is coarsely estimated and with estimated motion parameters,obvious geometry distortion is removed from each IR to obtain its "coarse" template as well as prediction of IR in other image is obtained.Thirdly,with IR "coarse" template and its prediction in other image,rigid affine transformation model is sought though iterative hypothesis-validation procedure and based on the model,"fine" template ensuring a locally maximum NCC computation is generated for each IR as well as its conjunctive is determined.Finally,the experimental results over different view/source of stereo image show that the proposed approach is robust to complicated image motion or geometry distortion.