融合匹配特征的立体图像颜色校正方法  

Stereoscopic Image Color Correction Based on Matching Features Fusion

作　　者：朱文婧 范媛媛 陈羽中[1,2] ZHU Wenjing;FAN Yuanyuan;CHEN Yuzhong(College of Computer and Data Science,Fuzhou University,Fuzhou 350105,China;Key Laboratory of Spatial Data Mining&Information Sharing,Ministry of Education,Fuzhou 350105,China)

机构地区：[1]福州大学计算机与大数据学院,福州350105 [2]空间数据挖掘与信息共享省部共建教育部重点实验室,福州350105

出　　处：《小型微型计算机系统》2024年第1期230-240,共11页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61972097)资助;福建省自然科学基金重点项目(2019J02006)资助;福建省高校产学合作项目(2018H6010)资助;福建省自然科学基金面上项目(2020J001494)资助.

摘　　要：立体图像颜色校正方法的目标在于消除立体图像左右视图之间的颜色差异.现有的立体图像颜色校正方法存在校正效果和时间效率不平衡的问题.为了解决这一问题,本文提出一种融合匹配特征的立体图像颜色校正方法.首先,通过视差注意力颜色校正网络得到待校正目标图像的初始校正图.然后,将初始校正图和参考图像输入基于光流的图像匹配网络得到光流匹配目标图.最后,由图像融合网络融合初始校正图、光流匹配目标图、参考图像和目标图像的特征并进行图像重建,得到最终的校正结果.实验结果表明,本文的方法具有先进的性能,能够在保持高时间效率的同时实现高质量的立体图像颜色校正效果.Stereoscopic image color correction aims to eliminate the color difference between the left and right views of the stereoscopic image.Existing stereo image color correction methods have problems in balancing correction effect and time efficiency.To solve this problem,we proposed a stereoscopic image color correction method based on matching features fusion.First,the parallax attention color correction network is used to obtain the initial correction result.Then,the initial correction result and the reference image are input into the optical flow-based image matching network to obtain the matching target image.Finally,fusing the initial correction result,the matching target image,the reference image,and the target image to obtain the final correction result through the image fusion network.Experiments show that the method proposed in this paper has advanced performance and can achieve high-quality stereoscopic image color correction while maintaining high time efficiency.

关 键 词：颜色校正 立体图像 卷积神经网络 图像融合 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]