关键词: 图像配准, 均匀化特征, 局部特征, 非极值抑制

Abstract:

A study aiming to achieve both the speed and accuracy of feature based multi-spectral image registration process is made. Then a new homogenized feature based image registration method is proposed. The method first uses the DOH operator to detect features, and then a circular filter window is used to adaptively select features based on the spatial distribution and intensity while keeping the features' rotation invariance. The selected features are further described and matched. Experiments show that the selected features are more robust and more uniformly distributed and that the method has great advantages in both speed and accuracy.

Key words: image registration, homogenized feature, local feature, non-maximum suppression