Scholarlink

<p>This study introduces a novel Gaussian curvature prior-based variational model for accurately estimating air light and depth maps, ultimately enabling the restoration of realistic haze-free, fog-free, or smoke-free images. The proposed work offers the following contributions: (i) air-light estimation and depth map estimation are first obtained using the dark channel prior as an initial guess. A Gaussian curvature regularizer is then applied to refine the estimates of air light transmission and the depth map. (ii) The method aims to recover an image that is free from haze, fog, or smoke. (iii) It successfully preserves the significant features of the depth map while simultaneously restoring the image. We use a synthesized data set of visual degradation scenarios from various fog, haze, and smoke conditions as well as the corresponding transmission maps to train the model in both indoor and outdoor environments. For the minimization and numerical solution of the proposed functional, an augmented Lagrangian method and fast Fourier transform method are applied. The empirically fixed Lagrange multipliers are tuned and adjusted automatically during each iteration according to the scene. The computational complexity of the proposed method/algorithm in terms of scalability and real-time application is also studied. We performed various experiments on images of the reside data set (<a href="http://tinyurl.com/y7keuhvx">http://tinyurl.com/y7keuhvx</a>), the multi real-world foggy image data-set (MRFID, <a href="http://www.vistalab.ac.cn/MRFID-for-defogging/">http://www.vistalab.ac.cn/MRFID-for-defogging/</a>), hazy city data-set (IEEE data-port) and many other hazy data-sets including indoor as well as outdoor real and synthetic images. Experimental results demonstrate that the suggested method can produce high-quality outcomes at a substantially lower computing cost than recent methods.</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

Depth map estimation and single image de-hazing via Gaussian curvature prior

  • 期刊论文

  • 发表日期：

    2025年9月15日

查看全文
  • Asmat Ullah,
  • HongGuang Sun,
  • Muhammad Arif

摘要

This study introduces a novel Gaussian curvature prior-based variational model for accurately estimating air light and depth maps, ultimately enabling the restoration of realistic haze-free, fog-free, or smoke-free images. The proposed work offers the following contributions: (i) air-light estimation and depth map estimation are first obtained using the dark channel prior as an initial guess. A Gaussian curvature regularizer is then applied to refine the estimates of air light transmission and the depth map. (ii) The method aims to recover an image that is free from haze, fog, or smoke. (iii) It successfully preserves the significant features of the depth map while simultaneously restoring the image. We use a synthesized data set of visual degradation scenarios from various fog, haze, and smoke conditions as well as the corresponding transmission maps to train the model in both indoor and outdoor environments. For the minimization and numerical solution of the proposed functional, an augmented Lagrangian method and fast Fourier transform method are applied. The empirically fixed Lagrange multipliers are tuned and adjusted automatically during each iteration according to the scene. The computational complexity of the proposed method/algorithm in terms of scalability and real-time application is also studied. We performed various experiments on images of the reside data set (http://tinyurl.com/y7keuhvx), the multi real-world foggy image data-set (MRFID, http://www.vistalab.ac.cn/MRFID-for-defogging/), hazy city data-set (IEEE data-port) and many other hazy data-sets including indoor as well as outdoor real and synthetic images. Experimental results demonstrate that the suggested method can produce high-quality outcomes at a substantially lower computing cost than recent methods.