光子学报 ›› 2020, Vol. 49 ›› Issue (10): 1010002-1010002.doi: 10.3788/gzxb20204910.1010002

• 图像处理 • 上一篇    下一篇

多尺度卷积结合自适应双区间均衡化的图像增强

路皓翔1(), 刘振丙1(), 郭棚跃2, 潘细朋1   

  1. 1.桂林电子科技大学 计算机与信息安全学院,广西 桂林 541004
    2.桂林电子科技大学 电子工程与自动化学院,广西 桂林 541004
  • 收稿日期:2020-06-30 接受日期:2020-08-28 出版日期:2020-10-25 发布日期:2020-10-13
  • 作者简介:路皓翔(1991—),男,博士研究生,主要研究方向为图像处理、机器学习. Email: 646510477@qq.com|刘振丙(1980—),男,教授,博士,主要研究方向为图像处理、深度学习. Email: zbliu@guet.edu.cn
  • 基金资助:
    国家自然科学基金(61866009);科技部国家重点研发项目(2018AAA0102600);广西高校中青年教师科研基础能力提升项目(2020KY05034);桂林电子科技大学科学研究基金(UF20001Y)

Multi-scale Convolution Combined with Adaptive Bi-interval Equalization for Image Enhancement

Hao-xiang LU1(), Zhen-bing LIU1(), Peng-yue GUO2, Xi-peng PAN1   

  1. 1.School of Computer and Information Security,Guilin University of Electronic Technology,Guilin,Guangxi 541004,China
    2.School of Electronic Engineering and Automation,Guilin University of Electronic Technology,Guilin,Guangxi 541004,China
  • Received:2020-06-30 Accepted:2020-08-28 Online:2020-10-25 Published:2020-10-13
  • Supported by:
    National Natural Science Foundation of China(61866009);Ministry of Science and Technology Research and Development of the National Key Project(2018AAA0102600);Guangxi University Young and Middle-aged Teachers' Research Ability Improvement Project(2020KY05034);Science Research Foundation of Guilin University of Electronic technology(UF20001Y)

摘要:

为了解决红外图像对比度低、细节模糊的问题,提出了多尺度卷积结合双区间自适应亮度均衡化的红外图像增强方法.首先采用多尺度卷积对图像进行预处理;然后以最大化类内方差且最小化类间方差作为遗传算法适应度函数求解图像亮暗图层的划分阈值,并采用引入细节信息的双区间直方图进行均衡化,同时通过引入均方差和均值的灰度均匀化方式提高图像亮度;最后,将自适应受限拉普拉斯算子提取的细节图像与亮度提升的图像进行线性加权融合重构出细节边缘清晰、对比度较强的图像.采用不同场景下红外图像和细节丰富的灰度图像进行试验并与传统方法进行对比来验证该方法的有效性.本文方法处理后的图像信息熵(Entropy, En)、熵增强(Enhancement by Entropy, EME)和平均梯度(Average Gradient, AG)最大增幅分别由原来的5.039 1、13.446 1和7.845 0增加到7.163 3、90.252 5和53.617 7 ,表明该方法具有更好的性能.

关键词: 图像增强, 对比度增强, 直方图均衡化, 红外图像, 细节锐化, 多尺度卷积

Abstract:

In order to solve the problem of low contrast and fuzzy details of infrared image, an infrared image enhancement method based on multi-scale convolution combined with adaptive bi-interval luminance equalization is proposed. Firstly, the image is pre-processed by multi-scale convolution; Then, the threshold of image segmentation is solved by genetic algorithm, where the function of maximizing intra class variance and minimizing inter class variance is taken as its fitness function, the double interval histogram with detail information is used to equalize, and the brightness of image is improved by introducing the gray level homogenization of mean square and mean square. Finally, the image with clear details and strong contrast is reconstructed by linear weighted fusion of the detail image extracted by adaptive limited Laplace and the image with brightness enhancement. Compared with conventional methods in the infrared images of different scenes and gray images with abundant details to verify the validity of the proposed method, the maximum growth rates of En, EME and AG in images processed by this method increased from 5.039 1, 13.446 1 and 7.845 0 to 7.163 3, 90.252 5 and 53.617 7, respectively. The experimental results show that this method has better performance.

Key words: Image enhancement, Contrast enhancement, Histogram equalization, Infrared image, Detail sharpening, Multi-scale convolution

中图分类号: