摘要
针对动平台视频序列图像配准中的特征点误匹配问题,提出了一种基于马尔可夫随机场和尺度不变特征变换(MRF-SIFT)的动平台视频序列图像配准算法。该方法利用SIFT特征点提取和改进的随机抽样一致性(RANSAC)剪枝进行特征点初匹配,然后在马尔可夫随机场模型框架下引入特征点的局部几何约束信息,通过求解匹配模型的最大后验概率估计寻优最佳匹配特征点,最后通过最小平方差方法计算匹配图像之间的变换矩阵。实验结果表明该方法在不同的动平台下能够更好地实现对视频序列图像的配准。
To deal with the miss matching of feature points on moving platform in video registration,a new algorithm based on MRF-SIFT( Markov Random Field- Scale Invariant Feature Transform) was proposed. SIFT feature points extraction and improved RANdom SAmple Consensus( RANSAC) algorithm was used for coarse matching. Then,geometric information among local feature points was encoded into the Markov random field framework to enhance the matching consistency of pixels. Finally,transform matrix was computed by least square method,which was used to realize video registration. Experimental results for different sequences show the proposed method has a better veracity.
出处
《计算机应用》
CSCD
北大核心
2015年第A02期230-233,共4页
journal of Computer Applications
关键词
视频序列
图像配准
尺度不变特征变换
马尔可夫随机场
特征点
video sequence
image registration
Scale Invariant Feature Transform(SIFT)
Markov Random Field(MRF)
feature point