Scholarlink

<p>To address the limitations of existing underwater image processing algorithms in turbid environments with single prior conditions, over-enhancement, low brightness and color distortion, this paper proposes a turbid image restoration method based on the turbidity evaluator model to optimize the local background light and transmission map. By integrating multivariate Gaussian model with underwater image quality assessment metrics, we establish a turbidity evaluator model to refine the prior local background light and transmission map to ensure that the turbidity estimation result of the restored image approximates that of clear in-air images. The optimization employs MADGRAD and AdaBelief optimizers. Then the image is restored based on underwater imaging model. The approach mitigates over-processing in low-brightness regions by leveraging local background light, and avoids inappropriate handling caused by single prior condition through optimization steps. Experimental results demonstrate that the proposed algorithm can effectively reduce turbidity and significantly improve image quality. The average UIQM and UCIQE values of the restored images reach approximately 3.2 and 0.63, ranking among the top-performing methods when compared to other enhancement and restoration algorithms like ULAP and UGAN. In object recognition tasks, the proposed algorithm achieves significantly improvement for object detecting, with an average recognition confidence score surpassing other algorithms benchmarked in this paper.</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

Turbid image restoration based on optimization via turbidity evaluator model

  • 期刊论文

  • 发表日期：

    2025年11月20日

查看全文
  • Chaoqi Chen,
  • Xiuzhan Zhang,
  • Haosu Zhang,
  • Jiancheng Liu,
  • Hao Li

摘要

To address the limitations of existing underwater image processing algorithms in turbid environments with single prior conditions, over-enhancement, low brightness and color distortion, this paper proposes a turbid image restoration method based on the turbidity evaluator model to optimize the local background light and transmission map. By integrating multivariate Gaussian model with underwater image quality assessment metrics, we establish a turbidity evaluator model to refine the prior local background light and transmission map to ensure that the turbidity estimation result of the restored image approximates that of clear in-air images. The optimization employs MADGRAD and AdaBelief optimizers. Then the image is restored based on underwater imaging model. The approach mitigates over-processing in low-brightness regions by leveraging local background light, and avoids inappropriate handling caused by single prior condition through optimization steps. Experimental results demonstrate that the proposed algorithm can effectively reduce turbidity and significantly improve image quality. The average UIQM and UCIQE values of the restored images reach approximately 3.2 and 0.63, ranking among the top-performing methods when compared to other enhancement and restoration algorithms like ULAP and UGAN. In object recognition tasks, the proposed algorithm achieves significantly improvement for object detecting, with an average recognition confidence score surpassing other algorithms benchmarked in this paper.