Scholarlink

<p>The process of haze removal has developed into a crucial pre-processing step for various applications, including automatic number plate recognition (ANPR) and object detection in advanced driver-assistance systems (ADAS). This paper presents a rapid and effective approach for haze removal based on the dark channel prior (DCP) method and the atmospheric scattering model. The proposed method aims to decrease the computational time of the DCP-based approach by breaking the process flow dependency between the estimation of the transmission map and the atmospheric light, thereby facilitating the concurrent processing of these two stages. The approach incorporates dynamic air-light estimation and the color channel transfer (CCT) technique to improve performance. The performance of the proposed method is assessed using both full-reference and no-reference performance metrics. The proposed method demonstrated a 21.9% increase in PSNR and achieved an improved structural similarity index (SSIM) of approximately 0.797 and a feature-retained feature similarity index (FRFSIM) of 0.6966 when evaluated using publicly available datasets. Specifically, synthetic datasets with varying haze conditions, such as O-HAZE, DENSE HAZE, and NH-HAZE, along with real-world datasets including RW-HAZE and RESIDE, were used to benchmark the method across diverse imaging scenarios. In addition to peak signal-to-noise ratio (PSNR), SSIM, and FRFSIM, perceptual quality was evaluated using natural image quality evaluator (NIQE) and blind/referenceless image spatial quality evaluator (BRISQUE), confirming robustness under both synthetic and natural haze conditions. The study also explored multiple transmission map refinement techniques and evaluated post-processing with a warmth filter, conducting an ablation analysis to optimize the proposed approach. Runtime and computational complexity were analyzed, with the proposed method achieving a significant reduction in FLOPS compared to SOTA approaches, including deep learning-based approaches, enabling real-time performance and competitive performance in both image quality and efficiency.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Concurrent processing in DCP-based image dehazing with color channel transfer

  • 期刊论文

  • 发表日期：

    2026年5月22日

查看全文
  • K Vidyamol,
  • M Surya Prakash,
  • Praveen Sankaran

摘要

The process of haze removal has developed into a crucial pre-processing step for various applications, including automatic number plate recognition (ANPR) and object detection in advanced driver-assistance systems (ADAS). This paper presents a rapid and effective approach for haze removal based on the dark channel prior (DCP) method and the atmospheric scattering model. The proposed method aims to decrease the computational time of the DCP-based approach by breaking the process flow dependency between the estimation of the transmission map and the atmospheric light, thereby facilitating the concurrent processing of these two stages. The approach incorporates dynamic air-light estimation and the color channel transfer (CCT) technique to improve performance. The performance of the proposed method is assessed using both full-reference and no-reference performance metrics. The proposed method demonstrated a 21.9% increase in PSNR and achieved an improved structural similarity index (SSIM) of approximately 0.797 and a feature-retained feature similarity index (FRFSIM) of 0.6966 when evaluated using publicly available datasets. Specifically, synthetic datasets with varying haze conditions, such as O-HAZE, DENSE HAZE, and NH-HAZE, along with real-world datasets including RW-HAZE and RESIDE, were used to benchmark the method across diverse imaging scenarios. In addition to peak signal-to-noise ratio (PSNR), SSIM, and FRFSIM, perceptual quality was evaluated using natural image quality evaluator (NIQE) and blind/referenceless image spatial quality evaluator (BRISQUE), confirming robustness under both synthetic and natural haze conditions. The study also explored multiple transmission map refinement techniques and evaluated post-processing with a warmth filter, conducting an ablation analysis to optimize the proposed approach. Runtime and computational complexity were analyzed, with the proposed method achieving a significant reduction in FLOPS compared to SOTA approaches, including deep learning-based approaches, enabling real-time performance and competitive performance in both image quality and efficiency.