Haze Removal via Regional Saturation-Value Translation and Soft Segmentation

TL;DR

This work introduces Regional Saturation-Value Translation (RSVT), a dehazing prior designed to mitigate color distortion in bright regions by exploiting minimal hue change and a clustering of S-V lines near atmospheric light in HSV space. It pairs RSVT with a novel soft segmentation built on a morphological min-max channel, and integrates RSVT with the Dark Channel Prior (DCP) to form a robust single-image dehazing framework. The method estimates atmospheric light from middle-ground regions, computes a per-pixel translation along saturation-value coordinates, and recovers a haze-free image through a convex combination that blends RSVT and DCP outputs. Across synthetic and real hazy datasets, RSVT achieves superior or competitive performance against classic priors and several lightweight DL models, while offering a practical runtime on CPU and improved sky-region restoration.

Abstract

This paper proposes a single image dehazing prior, called Regional Saturation-Value Translation (RSVT), to tackle the color distortion problems caused by conventional dehazing approaches in bright regions. The RSVT prior is developed based on two key observations regarding the relationship between hazy and haze-free points in the HSV color space. First, the hue component shows marginal variation between corresponding hazy and haze-free points, consolidating a hypothesis that the pixel value variability induced by haze primarily occurs in the saturation and value spaces. Second, in the 2D saturation-value coordinate system, most lines passing through hazy-clean point pairs are likely to intersect near the atmospheric light coordinates. Accordingly, haze removal for the bright regions can be performed by properly translating saturation-value coordinates. In addition, an effective soft segmentation method based on a morphological min-max channel is introduced. By combining the soft segmentation mask with the RSVT prior, a comprehensive single image dehazing framework is devised. Experimental results on various synthetic and realistic hazy image datasets demonstrate that the proposed scheme successfully addresses color distortion issues and restores visually appealing images. The code of this work is available at https://github.com/tranleanh/rsvt.

Figures (16)

• Figure 1: Dehazing results of various approaches on hazy images containing bright regions: (a) input, (b) DCP he2010single, (c) CEP bui2017single, (d) NLID berman2016non, (e) the proposed RSVT method, and (f) clean image.
• Figure 2: Regional Saturation-Value Transition: (a) hazy image, (b) sky regions in hazy (top) and clean (bottom) images marked in red color, (c,d) hazy (blue) and clean (red) points for the sky region in HSV color space from Hue-Value and Saturation-Value perspectives, respectively (orange segments connect corresponding hazy-clean pairs), (e) clean image, and (f) dehazed image by the proposed method.
• Figure 3: Statistical analyses: (a) differences between the hue components of hazy and haze-free point pairs within the sky regions (each bin stands for 10 distance levels), (b) distances between the intersections of S-V lines and the global atmospheric light coordinates in the Saturation-Value coordinate system (each bin stands for 10 distance levels), and (c) distribution of the correlation between the S-V ratio and the transmission.
• Figure 4: Morphological min-max channel: (a) hazy image, (b) min channel with a large radius, (d) min channel with a small radius, (f) max channel with a large radius, and (c,e,g) refined images of (b,d,f), respectively.
• Figure 5: Fused channel: (a) hazy image, (b) refined min channel, (c) refined max channel, and (d) fused channel of (b) and (c) through spatial-wise multiplication. The red and green boxes indicate oppositely low-intensity and high-intensity image patches from two channels at the same location, respectively.