基于稀疏性非负矩阵分解的偏振图像快速融合方法

大气与环境光学学报 ›› 2014, Vol. 9 ›› Issue (3): 229-236.

基于稀疏性非负矩阵分解的偏振图像快速融合方法

曾献芳1，徐国明2，易维宁3，黄红莲3，尹成亮2

（1. 安徽水利水电职业技术学院，安徽合肥，230601；
2. 解放军陆军军官学院，安徽合肥，230031；
3. 中国科学院安徽光学精密机械研究所，安徽合肥，230031）

收稿日期:2013-09-30 修回日期:2013-11-25 出版日期:2014-05-28 发布日期:2014-05-21
通讯作者: 徐国明（1979-），男，安徽太和人，讲师，博士研究生，主要研究方向为稀疏表示、超分辨率重建。 E-mail:xgm121@163.com
作者简介:曾献芳（1973-），女，安徽凤台人，副教授，硕士，主要研究方向为数字电子技术、图像处理。
基金资助:
安徽省自然科学基金项目(1208085QF126)资助

Polarimetric Image Fast Fusion Method Via Sparse Non-Negative Matrix Factorization

ZENG Xian-fang1, XU Guo-ming2, YI Wei-ning3, HUANG Hong-lian3, YIN Cheng-liang2

( 1. Anhui Teehnieal College of Water Resources and Hydroelectric Power, Hefei 230601, China;
2. Army Officer Academy, PLA, Hefei 230031, China;
3. Anhui Institute of Optics and Fine Mechanics, Chinese Academy of Sciences, Hefei 230031, China)

Received:2013-09-30 Revised:2013-11-25 Published:2014-05-28 Online:2014-05-21

摘要/Abstract

摘要：

针对基于非负矩阵分解(non-negative matrix factorization, NMF)的偏振图像融合方法效率低的不足，提出一种基于稀疏性NMF的偏振图像快速融合方法。首先，以偏振信息解析得到的各偏振参量图像构造原始数据集，其次，对NMF增加稀疏性约束，利用稀疏表示下的在线字典学习算法进行快速分解，然后对分解得到的三幅特征基图像按清晰度和方差进行排序，将排序后的特征基图像经直方图匹配及HSI颜色映射后，变换到RGB颜色空间，得到融合图像。与基于NMF的方法相比，运行时间提高约120倍，达到约1.5 s完成一次融合过程。实验结果验证了该方法在改善融合效果的同时，运行效率明显提高。

关键词: 图像融合, 偏振图像, 在线字典学习, 稀疏性非负矩阵分解

Abstract:

To improve the efficiency of the polarimetric image fusion methods via NMF, a fast fusion method based on sparse NMF was proposed. Firstly, the polarization parameter images were acquired by computing from intensity images of different polarization angle. The original data set was organized by the polarization parameter images. Secondly, the sparse constraint was added to NMF and the cost function was solved by online dictionary learning algorithm of sparse representation. Then, the data set was factorized by sparse NMF and three feature basis images sorted by definition and variance were obtained. Next, after histogram matching, these three sorted feature basis images were mapped into three color channels of HSI color model. Finally, the fused image was achieved by transforming the image from HSI to RGB color model. Compared with the method based on NMF, the running time is improved with 120 times. One fusing process can be finished in 1.5 s. Experiment results show that the proposed method not only has good fusion results but also enhances the running efficiency evidently.

Key words: image fusion, polarimetric image, online dictionary learning, sparse non-negative matrix factorization

中图分类号:

TN911.73

曾献芳徐国明易维宁黄红莲尹成亮. 基于稀疏性非负矩阵分解的偏振图像快速融合方法[J]. 大气与环境光学学报, 2014, 9(3): 229-236.

ZENG Xian-Fang, XU Guo-Ming, YI Wei-Ning, HUANG Hong-Lian, YIN Cheng-Liang. Polarimetric Image Fast Fusion Method Via Sparse Non-Negative Matrix Factorization[J]. Journal of Atmospheric and Environmental Optics, 2014, 9(3): 229-236.

参考文献

[1] Lee D D, Seung H S. Learning the parts of objects by non-negative matrix factorization [J]. Nature, 1999, 401(21): 788-791.
[2] Lee D D, Seung H S. Algorithms for non-negative matrix factorization [C]. In Advances in Neural Information Processing Systems, 2001: 556-562.
[3] Miao Qiguang, Wang Baoshu. A novel algorithm of multi-sensor image fusion using non-negative matrix factorization [J]. Journal of Computer-Aided Design & Computer Graphics, 2005, 17(9): 2029-2032(in Chinese).
苗启广, 王宝树. 图像融合的非负矩阵分解算法[J].计算机辅助设计与图形学学报, 2005,17(9): 2029-2032.
[4] Miao Qiguang, Wang Baoshu. Multi-focus image fusion based on non-negative matrix factorization [J]. Acta Optica Sinica, 2005, 25(6) : 755-759(in Chinese).
苗启广, 王宝树. 基于非负矩阵分解的多聚焦图像融合研究 [J]. 光学学报, 2005, 25 (6) : 755-759.
[5] Li Le, Zhang Yujin. A survey on algorithms of non-negative matrix factorization [J]. Acta Electronica Sinica, 2008, 36(4): 737-743(in Chinese).
李乐, 章毓晋. 非负矩阵分解算法综述 [J]. 电子学报, 2008, 36(4): 737-743.
[6] Zhou Pucheng, Han Yusheng, Xue Mogen, et al. Polarization image fusion approach based on non-begative matrix factorization and IHS color model [J]. Acta Photonica Sinica, 2010, 39(9): 1682-1687(in Chinese).
周浦城, 韩裕生, 薛模根, 等. 基于非负矩阵分解和IHS 颜色模型的偏振图像融合方法 [J]. 光子学报, 2010, 39(9): 1682-1687.
[7] Shi J, Ren X, Dai G, Wang J, et al. A non-convex relaxation approach to sparse dictionary learning [C]. CVPR, 2011: 1809-1816.
[8] Hoyer P O. Non-negative matrix factorization with sparseness constraints [J]. J. Mach. Learn. Res., 2004, 5(11): 1457-1469.
[9] Cai Lei, Zhu Yongsheng. Time-frequency spectra recognition based on sparse non-negative matrix factorization and support vector machine [J]. Acta Automatica Sinica, 2009, 35(10): 1272-1277(in Chinese).
蔡蕾, 朱永生. 基于稀疏性非负矩阵分解和支持向量机的时频图像识别 [J]. 自动化学报, 2009, 35(10): 1272-1277.
[10] Mairal J, Bach F, Ponce J, et al. Online learning for matrix factorization and sparse coding [J]. J. Mach. Learn. Res., 2010, (11): 19-60.
[11] Lu C W, Shi J P, Jia J Y. Online robust dictionary learning [C]. CVPR, 2013: 414-422.
[12] Osborne M R, Presnell B, Turlach B A. A new approach to variable selection in least squares problems [J]. IMA J. Numer. Anal., 2000, 20(3): 389-403.
[13] Tseng P. Convergence of a block coordinate descent method for non-differentiable minimization [J]. J. Optim. Theory Appl., 2001, 109(3): 475-494.
[14] Du Anping, Zhao Yongqiang, Pan Quan, et al. Image enhancement algorithm based on polarization character [J]. Computer Measurement & Control, 2007, 15(1): 106-108(in Chinese).
都安平, 赵永强, 潘泉, 等. 基于偏振特征的图像增强算法 [J]. 计算机测量与控制, 2007, 15(1): 106-108.
[15] You Jing, Liu Xiao, Yi Weining, et al. A white balance method for multi-spectral color synthesis image [J]. Journal of Atmospheric and Environmental Optics, 2012, 7(4): 282-290(in Chinese).
游晶, 刘晓, 易维宁, 等. 一种处理多光谱彩色合成图像的白平衡法 [J]. 大气与环境光学学报, 2012, 7(4): 282-290.