基于双颜色空间与多网络融合的水下图像增强  

Underwater image enhancement based on dual color space and multi⁃network fusion

作　　者：宋宇康 唐贵进[1] SONG Yukang;TANG Guijin(Jiangsu Key Laboratory of Image Processing and Image Communication,Nanjing University of Posts and Telecommunications,Nanjing 210003,China)

机构地区：[1]南京邮电大学江苏省图像处理与图像通信重点实验室,江苏南京210003

出　　处：《南京邮电大学学报（自然科学版）》2023年第3期44-56,共13页Journal of Nanjing University of Posts and Telecommunications：Natural Science Edition

基　　金：江苏省科协提升计划(TJ215039);南京邮电大学科研基金(NY219076)资助项目。

摘　　要：受不同水质、不同深度以及成像设备的影响,水下成像质量往往较低,有低对比度、颜色失真、细节模糊等问题。因此,文中在RGB颜色空间的基础上引入了HSV颜色空间,并与残差网络和注意力机制相结合,实现了双颜色空间与多个子网络的有机融合,弥补了单一颜色空间训练过程中信息丢失、特征提取不全的问题,在保证增强效果的同时提高了整个算法的泛化能力。实验在多个数据集上进行,与众多算法做主观视觉和客观评价指标的比较,使用全参考图像质量评价指标,均方误差(Mean Squared Error,MSE)、峰值信噪比(Peak Signal to Noise Ratio,PSNR)、结构相似性(Structural Similarity,SSIM)以及无参考水下图像质量评价指标,水下图像质量评价(Underwater Image Quality Measures,UIQM)等主流指标进行量化评价。实验表明,该算法不论是在主观还是客观评价上都具有一定的优越性。The quality of underwater images usually contain low contrast,color distortion,and blurred details,due to different factors like water quality,depths and imaging equipment.This paper introduces the HSV color space based on the RGB color space,and combines it with the residual network and the attention mechanism to realize the organic integration of the dual color space and multiple sub⁃networks.The proposed algorithm can alleviate the problems of information loss and incomplete feature extraction during the training process in a single color space.It not only ensures the enhancement effect but also improves the generalization ability.The experiments are carried out on multiple datasets,and compared the proposed algorithm with others in terms of subjective vision and objective evaluation.The objective evaluation includes the full reference image quality assessment indices such as mean squared error(MSE),peak signal to noise ratio(PSNR),and structural similarity(SSIM),as well as the no⁃reference underwater image quality assessment indices such as underwater image quality measures(UIQM).Experimental results show that our algorithm has an advantage in both subjective and objective evaluation.

关 键 词：水下图像 图像增强 卷积神经网络 颜色空间转换 注意力机制 

分 类 号：TN91[电子电信—通信与信息系统] TP183[电子电信—信息与通信工程]