经典的Harris特征点检测和Harris-Laplace特征点检测在传统的算法中占据重要的地位,但它们或在尺度或在冗余上仍存在问题,并且对弱的特征点不能很好地检测。为此,文章提出了一种改进的方法,在多尺度Harris检测特征点时用改进的双边滤波来替代传统的高斯滤波,而且在检测时对特征点进行分组,一组代表同一局部结构,然后在各组中选取一个自相关矩阵特征值最接近的点来代表这一结构。实验表明此方法能较好地定位特征点位置和去冗余特征点。
Harris and Harris-Laplace methods have the limitations of scales or redundant feature points, and can not solve the weak feature points, even though they have occupied an important posi- tion in the traditional algorithm. In this paper, an improved method of checking the feature points with multi-scale Harris method using an improved bilateral filter to replace the Gauss ~ter is presen- ted, then the tracking and grouping is conducted while checking the feature points, so some points re- presenting the same local structure are divided into a group, and a point which has the closest eigen- values of the auto-correlation matrix is selected from each group to represent its structure. The exper- imental results show that the method can precisely detect feature points and remove redundant feature points.