摘要
针对水下图像颜色失真和细节丢失等问题,提出了一种基于多尺度三重注意力机制的水下图像增强算法。该算法使用生成对抗网络作为基础架构,生成网络采用编码解码结构,并设计一个多尺度三重注意力模块,多尺度结构和三重注意力机制结合可以实现不同层次信息的跨维度交互,使网络更好地学习水下图像特征和抑制噪声特征,判别网络采用类似马尔可夫判别器的结构;构建了多个损失函数,使生成的图像在结构、内容、色彩上和参考图像保持一致。实验结果表明,所提算法在主观视觉和客观评价指标上均优于比较算法,可以有效地提升网络的特征提取能力,实现对不同场景水下图像的色彩恢复,增强图像的对比度和清晰度。
To solve the problems of color distortion and loss of details in underwater images,an underwater image enhancement algorithm based on Multi-Scale Triple Attention(MSTA)is proposed.The algorithm uses the Generative Adversarial Network(GAN)as the basic architecture,and the generative network adopts the encoding and decoding structure.An MSTA module is designed.The combination of the multi-scale structure and the Triple Attention(TA)mechanism can realize the cross-dimensional interaction of information at different levels,making the network better learn the features of underwater images and suppress the features of noise.The discriminant network adopts a structure similar to Markov discriminator.Multiple loss functions are constructed to make the generated image consistent with the reference image in terms of structure,content and color.The experimental results show that the proposed algorithm is superior to the comparison algorithms in terms of subjective visual effects and objective evaluation indicators.The proposed algorithm can effectively improve the feature extraction ability of the network,restore the color of underwater images in different scenes,and enhance the contrast and clarity of the images.
作者
陈海秀
陆康
何珊珊
黄仔洁
房威志
CHEN Haixiu;LU Kang;HE Shanshan;HUANG Zijie;FANG Weizhi(Nanjing University of Information Science&Technology,Nanjing 210000,China;Jiangsu Collaborative Innovation Center of Atmospheric Environment and Equipment Technology(CICAEET),Nanjing University of Information Science&Technology,Nanjing 210000,China)
出处
《电光与控制》
CSCD
北大核心
2023年第11期56-61,共6页
Electronics Optics & Control
基金
国家自然科学基金(61302189)
江苏省研究生科研与实践创新计划项目(SJCX23-0383)。
关键词
水下图像增强
三重注意力
生成对抗网络
编码解码结构
多尺度结构
underwater image enhancement
Triple Attention
Generative Adversarial Network
codec structure
multi-scale structure
