摘要
水下图像由于光在水中传播时的散射和吸收,而遭受色彩失真和低照度的影响。这些问题可能会干扰水下视觉任务,如识别和检测。根据传统的IFM,论文提出了一种协同学习神经网络来解决这些退化问题,该网络为两个子网T-net和A-net,它们分别用来估计传输图和环境光,通过协同学习的方式来恢复水下图像。实验证明,恢复的水下图像在几种最先进的方法中表现出更自然的颜色校正和更好的能见度提高,且拥有更好的稳定性、泛化性。
Underwater images suffer from color distortion and low illuminance due to scattering and absorption of light as it travels through water.These problems can interfere with underwater vision tasks such as recognition and detection.According to the traditional underwater optical image formation model,a collaborative learning neural network is proposed to solve these degradation problems.The network consists of two subnets,T-net and A-net,which are used to estimate the transmission image and the ambient light respectively,and restore the underwater image through collaborative learning.Experiments show that the restored underwater images show more natural color correction and better visibility improvement,and have better stability and generalization in sever-al state-of-the-art methods.
作者
王天语
陈苏婷
WANG Tianyu;CHEN Suting(School of Artificial Intelligence,Nanjing Univercity of Information Science&Technology,Nanjing 210044)
出处
《计算机与数字工程》
2025年第8期2319-2323,共5页
Computer & Digital Engineering
关键词
水下图像恢复
IFM
协同学习
数据合成
underwater image restoration
IFM
collaborative study
data synthesis
