Scale-based adaptive smoothing

A novel adaptive smoothing approach is proposed for noise removal and feature preservation where the local object scale is adopted to detect the size of the local structure. Its main idea is: the variance and mask size of the Gaussian filter are controlled by the local object scale. There are two main contributions in this paper. First one, we change a crucial parameter in the definition of object scale in previous reference in order to reflect the real size of object in visible light image very well. Second one, we introduce the median filter for removing the isolated large gradient noise points which interfere with the calculation of object scale greatly. Both subjective and objective comparisons demonstrate that our method can preserve the fine details while suppress noise, and is superior to the several traditional image smoothing algorithms.