测绘学报 ›› 2018, Vol. 47 ›› Issue (12): 1621-1629.doi: 10.11947/j.AGCS.2018.20170625

• 摄影测量与遥感 • 上一篇    下一篇

双范数混合约束的遥感影像亮度不均变分校正

李烁1,2, 王慧1, 耿则勋3, 于翔舟1, 卢兰鑫4   

  1. 1. 信息工程大学地理空间信息学院, 河南 郑州 450001;
    2. 61618部队, 北京 100094;
    3. 平顶山学院, 河南 平顶山 467000;
    4. 煤航信息产业有限公司, 陕西 西安 710054
  • 收稿日期:2017-11-16 修回日期:2018-09-18 出版日期:2018-12-20 发布日期:2018-12-24
  • 作者简介:李烁(1990-),男,博士生,研究方向为数字摄影测量和数字图像处理。E-mail:chxyls@163.com
  • 基金资助:
    国家自然科学基金(41571432)

Variational Uneven Illumination Correction with Double-norm Hybrid Constraints for Remote Sensing Imagery

LI Shuo1,2, WANG Hui1, GENG Zexun3, YU Xiangzhou1, LU Lanxin4   

  1. 1. Institute of Surveying and Mapping, Information Engineering University, Zhengzhou 450001, China;
    2. 61618 Troops, Beijing 100094, China;
    3. Pingdingshan University, Pingdingshan 467000, China;
    4. Aerial Photogrammetry and Remote Sensing Bureau, Xi'an 710054, China
  • Received:2017-11-16 Revised:2018-09-18 Online:2018-12-20 Published:2018-12-24
  • Supported by:
    The National Natural Science Foundation of China (No. 41571432)

摘要: 针对遥感影像内部亮度分布不均匀的问题,提出了一种双范数约束的亮度不均变分校正方法。首先分析了单范数变分Retinex方法的不足,根据反射影像和光照影像的特性构造新的变分模型,利用L1范数约束反射影像的纹理和细节信息,利用L2范数约束光照影像的光滑性。为了提高运算效率,采用分裂Bregman方法将变分模型的求解转换为3个子问题交替迭代计算。再对分离出的光照影像进行Gamma校正,取其均值与反射影像进行乘积运算,最后对重建影像进行对比度拉伸处理。试验结果表明,本文方法切实可行,与单范数约束的变分Retinex方法相比,本文方法可以更好地消除亮度不均匀现象和保持影像的纹理和细节信息,同时具有更高的运算效率,速度提高7倍以上。

关键词: 亮度不均, 变分校正, 双范数, 混合约束, 分裂Bregman

Abstract: In response to the uneven illumination of remote sensing image, a variational method of uneven illumination correction with double-norm hybrid constraints is proposed. Firstly, the deficiency of the variational method with single norm constraint is analyzed. A new variational model is constructed according to the characteristics of the reflectance image and the illumination image. The L1 norm and the L2 norm are adopted to constraint the texture and detail of reflectance image and the smoothness of the illumination image, respectively. Then, the solution of the variational model is translated into three sub-problems with split Bregman method and calculated using alternate iteration method to accelerate the computation. The mean of the illumination image enhanced with Gamma correction method is multiplied by the reflectance image. Finally, the contrast stretch transformation is implemented for a better performance. Experimental results indicate that the proposed method is practical and feasible. Compared with the variational Retinex method with single norm constraint, the proposed method performs better in correcting the uneven illumination and preserving the textures and details of the remote sensing image. In addition, the proposed method is more efficient, of which the running time is less than one seventh of that of the variational Retinex method with single norm constraint.

Key words: uneven illumination, variational correction, double-norm, hybrid constraints, split Bregman

中图分类号: