Optical Flow and Color Distribution-based Feature Point Matching
Kang Gewen · Modern Electronics Technique · 2009
To solve the difficulty of precise match between images,an optical flow registration and color distributions matching-based feature point matching algorithm which works well under large deformations and illumination changes are proposed.At first,a 5D spatial is defined and a group of parameters are employed to represent the information which including displacement,rotation,scale,the same parameters are used for transformation between the image patches.Then,these two matching algorithms can be combined into a single process which is called M-estimator for estimating the parameters of the transformation iteratively,a more accurate module is approaching to meet increasing accuracy request.The experiment proves that the feature points matching algorithm which combined optical flow with color distribution matching works well.Not only it has few cases of mistaken matches,but also it has high precision.