Atmospheric scattering caused by haze usually generates path radiance and degrades the quality of remotely sensed images. Traditionally, Dark Object Subtraction (DOS) is a commonly used approach to reduce such disadvantage on images. Nevertheless, when the distribution of haze is not spatially homogeneous, DOS may no longer applicable. Haze Optimized Transformation (HOT) defines a Haze Vector that can be used to estimate the haze condition for each particular image pixel. Accordingly, a correspondent gray value correction for reducing the influence of path radiance can be obtained as well. However, for images with higher spatial resolution, the results of HOT algorithm become spatially unstable because the complicated ground objects are more observable. In this study, for suppressing the drawbacks caused by detailed ground objects, a filtering process is introduced to smooth the original output of HOT. Furthermore, in order to decide a reasonable window size for medium filter, a high frequency noise analysis in frequency domain is also included. After the proposed method is tested by using high resolution satellite images, the results show that the haze removed image by proposed method can have better visual quality and applicability.