摘要
特征匹配是实现图像配准的重要手段,然而特征匹配中往往存在大量的误匹配,对于存在非刚性形变和大位移运动的图像序列尤为严重;如何从初始匹配结果中找到准确可靠的匹配点集,是提升图像配准性能的关键.为解决上述问题,首先根据图像的空间和色彩相似性,利用改进的超像素分割算法对图像进行分割;分割后的超像素块在空间上紧密相连,严格遵循图像轮廓边缘,且在同一区域内的颜色纹理基本趋于一致,可保证内部特征点具有相同或一致的运动趋势;同时,采用ORB算子对图像进行特征提取与描述,并利用暴力匹配算法得到初始匹配点集.其次,在超像素运动一致性约束下,提出了一种基于超像素运动统计模型的误匹配去除算法.通过建立超像素网格统计模型,将初始匹配坐标分配至相应的超像素区域,利用累加器计算出每个超像素对的匹配度,将初始匹配的概率分布特性转换为统计特性.最后,根据超像素匹配度的差异,计算出正确匹配的掩膜图像,实现了误匹配点的自动识别和剔除.仿真实验结果表明,与当前的误匹配去除算法相比,本文算法不依赖于复杂的参数模型,具有较高的鲁棒性,运算速度较快,可有效去除非刚性形变图像配准过程中产生的误匹配.
Feature matching is an important method for achieving image registration.However,a large number of mismatches always exist in feature matching,particularly for the image sequences with nonrigid deformations or large displacement motions.The key for improving image registration performance is by searching for accurate and reliable matching points from the initial matching results.To achieve this,an improved superpixel segmentation algorithm is first utilized to realize the image segmentation according to the space and color similarity of the images.The segmented superpixels are closely connected in space and the segmentation result follows the image contour.As the color and texture in the same region is consistent,the internal feature points tend to have the same or consistent motion trend.In addition,the oriented FAST and rotated BRIEF(ORB)operator is used to extract and describe the image feature points,and the initial matching point set is obtained using a brute-force matcher.Second,under the constraint of the superpixel motion consistency,a mismatch removal algorithm using the superpixel motion statistical model is proposed.By establishing the superpixel grid statistical model,the initial matching points are allocated to the corresponding superpixel regions.The matching score of each superpixel pair is calculated using the accumulator thereby transforming the probability distribution characteristics of the initial matching into the statistical one.Finally,the mask image of correct matching is calculated according to the difference in superpixel matching scores,and the automatic identification and removal of mismatches are realized.The experimental results show that compared with the state-of-the-art mismatch removal algorithm,the proposed algorithm does not rely on complex parameter models,has high robustness and fast calculation speed and can effectively remove the mismatches in nonrigid image matching.
作者
何凯
王阳
刘志国
马红悦
He Kai;Wang Yang;Liu Zhiguo;Ma Hongyue(School of Electrical and Information Engineering,Tianjin University,Tianjin 300072,China)
出处
《天津大学学报（自然科学与工程技术版）》
EI
CSCD
北大核心
2020年第2期147-153,共7页
Journal of Tianjin University：Science and Technology
基金
国家自然科学基金资助项目(61271326)~~
关键词
非刚性图像配准
误匹配去除
超像素运动统计
局部运动一致性
nonrigid image registration
mismatching removal
superpixel motion statistics
local motion consistency
