In this paper, we improved the original Haze-Optimized Transformation（HOT） method to solve the limitations of RGB synthetic images, such as sensitivity, over-correction, and color distortion. First, we combined the Normalized Difference Vegetation Index（NDVI）with the Red-Blue Spectral Difference（RBSD） to design a general mask that is insensitive to water bodies, bare soil, and man-made features.The mask was used to extract dense vegetation areas from the original image,and the corresponding regions of the initial HOT map were treated as a valid pixel set to assess haze intensity. The HOT values of invalid pixels based on the valid pixel set were inferred to generate a final valid HOT map. Finally, with the valid HOT map as reference to implement Dark Object Subtraction（DOS）, the influence of haze can be eliminated. The corrected value of the starting band was determined during DOS by calculating the percentile values of the histograms.The scattering model was then used to produce the correction values of other bands. The scatter plots of the blue and red bands before and after haze removal show that the hazed images share the same characteristics with the clear regions while maintaining differences between different objects. The plaques and halo artifacts are also significantly minimized. Furthermore, experimental results reveal that the improved HOT method can effectively remove haze and thin cloud, as well as resolve the limitations of synthetic images.

image dehazing;HOT transform;mask;dark object subtraction;scattering model