何凯, 闫佳星, 魏颖, 王阳
AuthorsHTML:何凯, 闫佳星, 魏颖, 王阳
AuthorsListE:He Kai, Yan Jiaxing, Wei Ying, Wang Yang
AuthorsHTMLE:He Kai, Yan Jiaxing, Wei Ying, Wang Yang
Unit:天津大学电气自动化与信息工程学院，天津 300072
Unit_EngLish:School of Electrical and Information Engineering, Tianjin University, Tianjin 300072, China
Abstract_Chinese:非刚性形变具有局部变形严重、非线性特征明显、各点转换模型不统一等特点, 其形变前后图像的自动配准一直是计算机视觉领域的研究热点和难点．为解决上述问题, 提出一种改进的光流场模型, 将特征向量守恒作为约束条件, 较好地解决了大位移形变配准问题, 同时保留了图像的细节信息, 提高了光流估计的准确性．此外, 为解决传统光流场模型中因过度平滑而导致的图像边缘模糊问题, 对金字塔各层位移场作了双边滤波, 以保证去除噪点的同时, 保留图像的边缘特征．实验结果表明, 该算法模型鲁棒性较高, 对于复杂形变的非刚性图像仍然能够获得理想的配准效果．
Abstract_English:Deformation of non-rigid objects has the characters of obvious local transformation，serious non-linearity and lacking of uniform transformation model. Due to these reasons，non-rigid image automatic registration before and after deformation is always a hot topic and difficulty in the computer vision field. To solve this challenging problem，an improved optical flow model is proposed using the conservative character vector as the constraint condition. Thus，the problem of large displacement registration is solved，the image detail is well maintained，then the accuracy of the estimated optical flow is improved. Moreover，to overcome the edge blur problem caused by the over-smoothness of traditional optical flow model，bilateral filtering is utilized on the displacement field of each level of pyramid to remove noise while preserving the image edge features. Experimental results demonstrate that the proposed model is robust and suitable to deal with the non-rigid images with complex deformation to achieve ideal registration effect.
Keyword_Chinese:图像配准; 非刚性形变; 光流场模型; 特征向量; 双边滤波
Keywords_English:image registration; non-rigid deformation; optical flow field model; character vector; bilateral filtering

PDF全文下载地址:http://xbzrb.tju.edu.cn/#/digest?ArticleID=6023