Sand Dust Images Enhancement Based on Red and Blue Channels

Fei Shi, Zhenhong Jia, Huicheng Lai, Sensen Song, Junnan WangSensors 2022 · DOI

TL;DR

This paper introduces a novel algorithm for enhancing sand dust images by leveraging the distinct properties of red and blue color channels. By first correcting color casts with a red-channel-based function and then removing dust particles with a blue-channel-based approach, the method significantly improves image clarity and visual quality, outperforming existing state-of-the-art restoration techniques.

What Problem Does This Paper Address?

Sandstorms severely degrade image quality, leading to low contrast, color casting (often yellowish), and loss of fine details. This degradation poses a significant challenge for computer vision systems operating in such environments, impacting applications like surveillance, traffic monitoring, and autonomous driving. Traditional image restoration methods often struggle to fully address these issues, particularly the persistent color casting and inaccurate estimation of scene transmission and atmospheric light. This paper aims to bridge this gap by developing a robust method specifically tailored for sand dust image enhancement.

Methods

The proposed algorithm comprises two key modules:

  • Red Channel Correction Function (RCC): This module addresses the color casting issue. The authors observe that red light is absorbed and scattered less by dust particles compared to blue and green light. The RCC module utilizes the red channel's histogram as a reference to adjust the blue and green channels, effectively correcting the yellowish or reddish hue. It involves translating the blue and green channel histograms based on the mean of the red channel and then stretching the colors in RGB space using a formula that considers channel means and variances.
  • Blue Channel Dust Particle Removal (BDPR): This module focuses on removing the dust particles and restoring image details. It's based on the characteristic that blue light is rapidly absorbed by dust particles, leading to lower intensity in the blue channel of sand dust images. The method adapts the Dark Channel Prior (DCP) approach, which is commonly used for haze removal. The BDPR module leverages the corrected image from the RCC module and the original sand dust image to estimate atmospheric light and transmission maps. Specifically, it uses a modified DCP prior that considers the reciprocal of the blue channel intensity to better estimate scene depth and atmospheric light. After estimating transmission, the image is restored using a formula derived from the atmospheric scattering model.

Finally, the output from the BDPR module, which might still be darker with some blue tones, is further enhanced for contrast and then fused with the output of the RCC module using wavelet fusion to produce the final clear and visible image.

Datasets

The paper does not explicitly mention training datasets. For evaluation, the authors used 12 sand dust images from Figures 6 and 7, and an additional 375 sand dust images collected from the internet. These images represent various sandstorm conditions, including weak color cast and severe dust.

Key Insights and Ideas

  • Exploiting Channel-Specific Dust Interaction: The core insight is that red and blue channels are affected differently by sand dust. Red light penetrates better, while blue light is absorbed more readily. This differential absorption is the foundation for both color correction and particle removal.
  • Synergistic RCC and BDPR: Combining a dedicated color correction module (RCC) with a dust removal module (BDPR) is crucial. The RCC module prepares the image for more accurate transmission estimation by the BDPR module, overcoming limitations of traditional methods that struggle with color cast.
  • Modified Dark Channel Prior for Dust: Adapting the well-established Dark Channel Prior (DCP) by focusing on the blue channel's behavior in dust-laden images allows for more effective estimation of transmission and atmospheric light, which are critical for restoration.
  • Improved Atmospheric Light Estimation: The proposed method selects atmospheric light from the most opaque sand dust areas, which is shown to be more effective than selecting the brightest sky areas, as done by some other DCP-based algorithms.

Key Results

  • Qualitative Improvement: The proposed method visually produces clearer images with reduced color casting and more visible details compared to 10 other state-of-the-art algorithms. Figures 6 and 7 demonstrate its effectiveness across various sand dust scenarios.
  • Quantitative Performance: Using non-reference evaluation metrics (visible edges recovery percentage 'e', saturation 'σ', contrast restoration percentage 'r', NIQE, DIIVINE, and NPQI), the proposed method achieved competitive or superior results. Notably, it achieved better scores on DIIVINE and NPQI for 375 sand dust images, indicating better restoration quality and perceptual characteristics.
  • Outperforming State-of-the-Art: The paper claims the method significantly improves image quality and outperforms existing restoration algorithms.

Limitations and Future Work

  • Running Time: The authors acknowledge that the proposed method has a higher time cost due to the use of guided filtering and local filtering. This limits its applicability in real-time systems.
  • Future Work:
    • Improving the efficiency of the algorithm to meet real-time requirements.
  • Developing methods for sand dust video restoration using spatio-temporal data modeling.

Summary updated

Abstract

The scattering and absorption of light results in the degradation of image in sandstorm scenes, it is vulnerable to issues such as color casting, low contrast and lost details, resulting in poor visual quality. In such circumstances, traditional image restoration methods cannot fully restore images owing to the persistence of color casting problems and the poor estimation of scene transmission maps and atmospheric light. To effectively correct color casting and enhance visibility for such sand dust images, we proposed a sand dust image enhancement algorithm using the red and blue channels, which consists of two modules: the red channel-based correction function (RCC) and blue channel-based dust particle removal (BDPR), the RCC module is used to correct color casting errors, and the BDPR module removes sand dust particles. After the dust image is processed by these two modules, a clear and visible image can be produced. The experimental results were analyzed qualitatively and quantitatively, and the results show that this method can significantly improve the image quality under sandstorm weather and outperform the state-of-the-art restoration algorithms.
BibTeX
@inproceedings{shi2022sand,
  author = {Fei Shi and Zhenhong Jia and Huicheng Lai and Sensen Song and Junnan Wang},
  title = {{Sand Dust Images Enhancement Based on Red and Blue Channels}},
  booktitle = {Sensors},
  year = {2022},
  doi = {10.3390/s22051918},
}
Figure