西安电子科技大学学报 ›› 2022, Vol. 49 ›› Issue (3): 137-146.doi: 10.19665/j.issn1001-2400.2022.03.016

• 计算机科学与技术&人工智能 • 上一篇    下一篇

融合显著性信息的水下图像清晰化算法

王朝宇(),郭继昌(),王天保(),郑司达(),张怡()   

  1. 天津大学 电气自动化与信息工程学院,天津 300072
  • 收稿日期:2021-01-31 修回日期:2021-12-08 出版日期:2022-06-20 发布日期:2022-07-04
  • 通讯作者: 郭继昌
  • 作者简介:王朝宇(1996—),女,天津大学硕士研究生,E-mail: wangzhaoyuisjoy@tju.edu.cn|王天保(1994—),男,天津大学硕士研究生,E-mail: wangtianbao@tju.edu.cn|郑司达(1995—),男,天津大学硕士研究生,E-mail: zhengsida@tju.edu.cn|张怡(1997—),男,天津大学硕士研究生,E-mail: zhangyi123@tju.edu.cn
  • 基金资助:
    国家自然科学基金(61771334)

Algorithm for clarification of the underwater image combining saliency information

WANG Zhaoyu(),GUO Jichang(),WANG Tianbao(),ZHENG Sida(),ZHANG Yi()   

  1. School of Electrical and Information Engineering,Tianjin University,Tianjin 300072,China
  • Received:2021-01-31 Revised:2021-12-08 Online:2022-06-20 Published:2022-07-04
  • Contact: Jichang GUO

摘要:

由于水对光的选择性吸收和水中微粒的散射效应,水下光学图像通常存在颜色失真、对比度低和细节模糊等缺陷。为解决水下图像存在的颜色失真和低对比度问题,提出了一种融合显著性信息的水下图像清晰化算法。首先,使用基于四叉树分割的分层搜索算法估计背景光,结合水下成像模型对水下图像进行初步清晰化;同时,进行简单线性迭代聚类超像素分割,并根据各超像素与边界背景聚类的特征相似度构建全局距离矩阵,再由多层元胞自动机整合生成显著图;最后,在Lab 颜色空间依据图像的显著性信息,对水下图像分区域进行颜色校正。选取UFO-120数据集中的1 500张水下图像进行实验,该算法在局部块对比度、熵、清晰度测量指标、对比度测量指标及主观颜色上有显著提升。实验结果表明,这种算法在水下图像颜色校正和对比度增强方面存在明显优势。

关键词: 水下图像清晰化, 颜色校正, 图像显著性, k均值聚类

Abstract:

Due to the selective absorption of light by water and the scattering effect of particles in water,underwater images usually have some defects,such as color distortion,low contrast and blurred details.Considering the color distortion and low contrast in the underwater images,an underwater image clarification algorithm combining saliency information is proposed.First,the background light is estimated by a hierarchical search algorithm based on quadtree segmentation.Second,in combination with the underwater imaging model,the perliminary clarification of the underwater image is performed.Furthermore,the superpixels are achieved via the Simple Linear Iterative Clutering algorithm.The global distance matrixes are constructed according to the feature similarity of each superpixel and the boundary background clusters.Then,the global distance matrixes are integrated to generate a saliency map by the Multi-Layer Cellular Automata.Finally,based on the saliency map,the color of underwater images is corrected in the Lab color space.In the experiment,1500 underwater images in UFO-120 dataset are selected as research objects.The algorithm has a significant improvement in the Patch-based Contrast Quality Index,Entropy,Underwater Image Sharpness Measure,Underwater Image Contrast Measure and subjective color restoration.Extensive experiments show that the proposed algorithm outperforms state-of-the-art methods in color correction and contrast enhancement of underwater images.

Key words: underwater image clarification, color correction, image saliency, k-means clustering

中图分类号: 

  • TP391.41