.png){kind=icon}
返回
An efficient learning-based method for underwater image enhancement
.png){kind=icon}
PRE
AI
DOI：10.1016/j.displa.2022.102174.png)
.png){kind=icon}
原文求助.png){kind=icon}
分享.png){kind=icon}
收藏
摘要
En 中文
The Underwater image enhancement, which targets removing the noise, scattering effect, and the bluish, greenish, or yellowish tone from underwater images, has lately attracted much attention. Traditional methods generally assume some prior knowledge when performing constrained optimization. Recently, the deep neural network (DNN)-based methods have exhibited superior performance over the traditional methods. However, state-of-the-art DNN models generally cost a large number of parameters, which is unfriendly to practical implementation. By contrast, this paper proposes a simple yet effective convolution neural network (CNN) based framework to enhance underwater images. Our method consists of two stages, CNN-based enhancement and YUV-based post-processing. In the first stage, a lightweight CNN network is developed to extract the latent features from the input image for enhancement. Specifically, our CNN cascades three residual groups which leverage channel attention blocks to strengthen the feature extraction capability. In the second stage, the output of our CNN model is transformed to the YUV color space where the luminance component is further corrected, improving the brightness of the whole image. Our model is verified on different underwater image datasets. Experimental results show that our method outperforms state-of-the-art methods both qualitatively and quantitatively. Moreover, in terms of computational complexity, our CNN model costs only 0.89M parameters which is only 21% and 81% of the existing method FUnIE-GAN and WaterNet, respectively.
Keyword:
Convolutional neural network
Underwater image enhancement
Channel attention
Color space
Evaluation metric
期刊
IF:
3.4
论文数:
2.3K
被引数:
3.2K
机构
引用论文
Unusual features of thyroid carcinomas in Japanese patients with Werner syndrome and possible genotype–phenotype relations to cell type and race.png)
Cancer
CancerIF0
PREAI
分享收藏A real-time image processor with combining dynamic contrast ratio enhancement and inverse gamma correction for PDPDISPLAYS
DISPLAYSIF3.4
PREAI
分享收藏Underwater image enhancement based on conditional generative adversarial network基于条件生成对抗网络的水下图像增强
PREAI
分享收藏A Parallel Processing Solid Phase Extraction Protocol for the Determination of Whole Blood Folate
PREAI
分享收藏Underwater Image Enhancement Using a Multiscale Dense Generative Adversarial Network使用多尺度密集生成对抗网络的水下图像增强
PREAI
分享收藏分享收藏
查看更多内容