Scholarlink

<p>Recent advancements in deep-sea exploration have highlight the demand for high quality underwater imaging, yet traditional cameras fails in light absorption and scattering. This leads to color distortion, low contrast, and blurring. To tackle these limitations, this work presents Generative Adversarial Network Markov-Patch and Conditional learning (MPC) with Generative Adversarial Network (GAN) which combines Markov-Patch learning within the GAN framework for Underwater Image Enhancement (UWIE). The suggested MPC-GAN incorporates Markov-Patch discrimination for ensuring spatial consistency by enforcing local feature coherence across neighboring patches, and overcoming unnatural artifacts. The generator follows a U-Net-based encoder-decoder structure with skip connections, preserve fine details and conditional learning manages the enhancement process. Extensive evaluations on the UFO120 underwater dataset demonstrate that MPC-GAN surpasses traditional and deep learning-based methods in terms of both perceptual quality and quantitative metrics. The suggested model achieves a PSNR of 59.51&#xa0;dB, MSE of 189.4, and SSIM of 93.91%. Qualitative outcomes show that MPC-GAN restores color accuracy, improves contrast, and enhances sharpness. These analyses establish MPC-GAN as a state-of-the-art solution for UWIE, bridging the gap between theoretical advancements and real-world deployment.</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

Enhancement of underwater images using novel Markov-patch based conditional generative adversarial network

  • 期刊论文

  • 发表日期：

    2025年7月25日

查看全文
  • K. T. Abdul Kader,
  • K. Siva Sankar

摘要

Recent advancements in deep-sea exploration have highlight the demand for high quality underwater imaging, yet traditional cameras fails in light absorption and scattering. This leads to color distortion, low contrast, and blurring. To tackle these limitations, this work presents Generative Adversarial Network Markov-Patch and Conditional learning (MPC) with Generative Adversarial Network (GAN) which combines Markov-Patch learning within the GAN framework for Underwater Image Enhancement (UWIE). The suggested MPC-GAN incorporates Markov-Patch discrimination for ensuring spatial consistency by enforcing local feature coherence across neighboring patches, and overcoming unnatural artifacts. The generator follows a U-Net-based encoder-decoder structure with skip connections, preserve fine details and conditional learning manages the enhancement process. Extensive evaluations on the UFO120 underwater dataset demonstrate that MPC-GAN surpasses traditional and deep learning-based methods in terms of both perceptual quality and quantitative metrics. The suggested model achieves a PSNR of 59.51 dB, MSE of 189.4, and SSIM of 93.91%. Qualitative outcomes show that MPC-GAN restores color accuracy, improves contrast, and enhances sharpness. These analyses establish MPC-GAN as a state-of-the-art solution for UWIE, bridging the gap between theoretical advancements and real-world deployment.