基于目标尺度的自适应加权函数特征点检测方法

为提高Harris特征点检测方法对噪声的鲁棒性,通过一般化目标尺度的概念,应用到检测算子的加权函数之中,使得对于不同噪声强度的图像,滤波模板具备自适应性.针对高斯函数和双边函数,在每一个像素通过邻域搜索的方式得到该像素的目标尺度,作为高斯函数的标准差和双边函数的空间标准差,进而可根据相关准则确定离散情况下滤波模板的大小.试验结果表明,各种检测算子的优化方案能够有效地滤除图像中的噪声,不但减少了将噪声作为角点的情况发生,而且对不同噪声的变化具备较好的鲁棒性.

The Method of Adaptive Weighting Function Feature Point Detection with the Object Scales

In order to improve the robustness of Harris feature detection methods with noise,by generalizing the principle of object scale,it is applied to modify the weighting functions of detecting operators.The adaptive filtering models can be obtained for different intensities.Through searching the optimal radius in the neighborhoods of a pixel,a object scale can be gotten,then for the standard deviations of Gaussian function and Bilateral function,a size of filtering model can also be computed in discrete circumstance.The experiments show that several proposed methods can filter the image noise better than before,not only reduce the wrong feature points from noise,but also have better robustness to the variable noise.