亚像素级模糊图像配准算法

针对目前的特征点匹配算法不能有效配准边缘模糊的图像,提出一种灰度偏移极值特征。将局部像素灰度偏移均值最明显的点作为特征点,利用最小二乘拟合曲线,通过曲线导函数将偏移极值点精确定位于亚像素级,最后采用分块的NMI为特征点分配32维特征向量,再对比向量间距离以实现配准。该特征易于提取、运算简单,而且能在阶跃边缘不清晰的图像中稳定存在。经实验验证,算法应用于模糊图像配准具有较高的匹配精度。

Fuzzy image registration algorithms of sub-pixel level

On account of the fact that today's point feature registration algorithms can't match edge fuzzy images effectively, this paper proposed the extreme gray offset feature. Firstly, defined the feature point as the pixel with the most dramatic offset from the local mean value. And then, located the extreme offset feature in sub-pixel level by derived function, while fitted the function curve by the least square. Finally, realized the registration by comparing the distance among those 32D feature vectors built by NMI. This feature was easy to extract and calculate, the results of experiment show that the algorithm has high matching accuracy applied in the fuzzy image registration.