Scholarlink

<p>Ocean research and engineering require the acquisition of underwater images to explore and interact with the marine environment. Transmission rates of these images are affected by low communication bandwidth. To address these problems, an end-to-end spatial-channel attention mechanism-based compression-reconstruction model using a shallow convolutional neural network (SC-SCNN) is proposed in the given paper. In this work, comparatively good quality reconstructed images are obtained using SC-CNN from very low resolution lossy compressed images. Experimental results show that the proposed method can obtain better Peak Signal to Noise Ratio (PSNR) values than the traditional compression techniques. The memory space consumed by the reconstructed images of the proposed method is less than the traditional techniques. The reconstructed images have better quality when compared to existing image restoration models such as super resolution CNN (SRCNN), efficient sub-pixel convolutional network (ESPCNN), and trainable nonlinear reaction–diffusion (TNRD).</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

Comprehensive compression and attention-based reconstruction model for enhancing underwater images

  • 期刊论文

  • 发表日期：

    2025年5月21日

查看全文
  • Rashmi S. Nair,
  • Ayalapogu Ratna Raju,
  • E. Ahila Devi,
  • Anusha Rajan,
  • Aarti Sangwan,
  • Nellore Manoj Kumar

摘要

Ocean research and engineering require the acquisition of underwater images to explore and interact with the marine environment. Transmission rates of these images are affected by low communication bandwidth. To address these problems, an end-to-end spatial-channel attention mechanism-based compression-reconstruction model using a shallow convolutional neural network (SC-SCNN) is proposed in the given paper. In this work, comparatively good quality reconstructed images are obtained using SC-CNN from very low resolution lossy compressed images. Experimental results show that the proposed method can obtain better Peak Signal to Noise Ratio (PSNR) values than the traditional compression techniques. The memory space consumed by the reconstructed images of the proposed method is less than the traditional techniques. The reconstructed images have better quality when compared to existing image restoration models such as super resolution CNN (SRCNN), efficient sub-pixel convolutional network (ESPCNN), and trainable nonlinear reaction–diffusion (TNRD).