Infrared and Visible Image Fusion Based on MSPCNN and FCM
-
摘要: 针对红外和可见光图像融合存在的轮廓信息不全、边缘及纹理细节信息缺失等问题,提出一种改进简化脉冲耦合神经网络(Improved Simplified Pulse Coupled Neural Network, MSPCNN)和模糊C-均值(Fuzzy C-mean, FCM)图像融合算法。首先,将红外和可见光图像用非下采样剪切波算法(Non-Subsampled Shearlet Transform,NSST)分解为高低频子带;然后对分解后的高频子带采用MSPCNN融合,用一种高斯分布权重矩阵进行处理,增强细节信息和对比度;接着,将得到的低频子带图像使用FCM聚类算法进行聚类中心提取,设置聚类中心近似阈值简化过程,实现背景分类提取;最后利NSST进行逆变换,从而完成红外和可见光的图像融合过程。通过客观评价指标计算,本文所提方法在平均梯度、标准差、平均相似度等参考指标上相对于其他同类型算法均有改善提高,由于模型参数的简化,算法运行速度相对于其他算法得到提升,算法更适用于复杂场景。Abstract: Aiming at the problems of incomplete contour information, missing edge and texture details in infrared and visible image fusion, A Improved Simplified Pulse Coupled Neural Network (MSPCNN) and Fuzzy C-mean (FCM) image fusion algorithm is proposed. First, infrared and visible images were decomposed into high and low frequency sub-bands using the Non-Subsampled Shearlet Transform (NSST).Then MSPCNN is used to fuse the decomposed high frequency subband, and a Gaussian distribution weight matrix is used for processing to enhance the detail information and contrast. Then, the obtained low-frequency sub-band images were extracted by using FCM clustering algorithm, and the approximate threshold of clustering center was set to simplify the process to achieve background classification extraction.Finally, the inverse transformation of NSST is carried out to complete the infrared and visible image fusion process.Through objective evaluation index calculation, compared with other algorithms of the same type, the method proposed in this paper has been improved in terms of average gradient, standard deviation, average similarity and other reference indexes. As the running speed of simplified algorithm of model parameters has been improved, the timeliness of the algorithm in this paper has been improved compared with other algorithms, and the algorithm is more suitable for complex scenarios.
-
表 1 不同方法的客观评价指标均值
Table 1. Objective evaluation index mean value of different methods
Image Algorithm AVG SSIM QAB/F PSN SF FD First PCA 3.4672 0.7171 0.4166 19.5587 8.1563 4.2357 NSCT-PCNN 2.6693 0.7345 0.3013 22.4713 5.8531 3.0063 NSST-PCNN 3.3602 0.7281 0.4503 19.5606 8.0915 4.0816 PCNN-IFS 3.3544 0.3994 0.5144 12.3504 8.0694 4.0632 NSST-FCM 4.8366 0.7226 0.4541 18.9521 12.6139 6.0441 NSST-PA-PCNN 4.3941 0.7196 0.4776 19.0621 11.0428 5.1949 Proposed algorithm 4.9783 0.7357 0.4673 19.5632 11.9764 5.8945 Second PCA 2.2481 0.6639 0.4406 12.7854 5.3698 2.7438 NSCT-PCNN 1.7512 0.6406 0.3292 18.6211 3.9054 1.9935 NSST-PCNN 2.4183 0.6599 0.4638 12.7745 5.8367 2.9133 PCNN-IFS 2.4097 0.4415 0.5566 11.9401 5.8081 2.8891 NSST-FCM 3.2193 0.6478 0.4973 14.7370 8.1806 4.3017 NSST-PA-PCNN 2.6837 0.6437 0.4906 12.6591 6.6781 3.2303 Proposed algorithm 3.3542 0.6546 0.5156 13.6431 8.3459 3.9543 Third PCA 3.1381 0.522 0.2667 16.6712 12.5317 5.5287 NSCT-PCNN 2.1691 0.5998 0.3712 18.7813 6.0813 2.4056 NSST-PCNN 2.6825 0.6185 0.4206 16.8078 9.7897 3.8225 PCNN-IFS 2.6759 0.4484 0.4881 16.7297 9.7805 3.8034 NSST-FCM 3.6972 0.6182 0.6442 17.2496 13.2931 4.6566 NSST-PA-PCNN 3.2976 0.6164 0.5943 16.4798 11.6122 4.0373 Proposed algorithm 3.6932 0.6245 0.6358 18.9682 12.3589 4.3589 Fourth PCA 2.7559 0.7871 0.5614 16.8755 7.7808 3.3122 NSCT-PCNN 1.5552 0.763 0.2586 16.9802 4.4557 1.8387 NSST-PCNN 2.7275 0.7872 0.6003 16.8603 9.1856 3.5813 PCNN-IFS 2.7188 0.4403 0.6549 12.5775 9.1723 3.5592 NSST-FCM 3.1382 0.7685 0.5462 17.0442 10.7773 4.0853 NSST-PA-PCNN 2.9165 0.7659 0.5529 16.7003 9.8139 3.6201 Proposed algorithm 3.3584 0.7756 0.5641 17.1298 9.9821 4.1269 Fifth PCA 4.9907 0.6288 0.4748 19.5355 9.9664 6.0269 NSCT-PCNN 3.0046 0.6506 0.2351 21.5392 5.9493 3.5833 NSST-PCNN 4.9841 0.6214 0.4186 19.4975 10.9044 6.5557 PCNN-IFS 4.9805 0.3504 0.4423 11.1055 10.8874 6.5449 NSST-FCM 6.2671 0.6559 0.4191 19.1803 13.5097 7.9851 NSST-PA-PCNN 5.3543 0.6463 0.4451 19.2727 11.4391 6.6561 Proposed algorithm 6.1596 0.7521 0.4563 19.6581 12.6985 6.7539 Sixth PCA 3.9552 0.5015 0.5236 19.7263 10.9258 4.1901 NSCT-PCNN 2.8944 0.4914 0.3599 21.4056 9.4389 2.9698 NSST-PCNN 3.5472 0.5064 0.5285 19.7583 10.5539 3.7219 PCNN-IFS 3.5657 0.4371 0.6022 16.0777 10.5725 3.7328 NSST-FCM 5.2416 0.5077 0.6883 18.5657 14.9875 5.5393 NSST-PA-PCNN 4.7394 0.5021 0.6371 19.0715 14.2598 4.9865 Proposed algorithm 5.3684 0.5129 0.6752 20.5691 13.5214 5.6321 表 2 不同算法在6种不同场景下的融合时间
Table 2. The fusion time of different algorithms in six different scenarios
s Algorithm Scenes of fusion image First Second Third Fourth Fifth Sixth PCA 4.603 1.515 4.744 3.646 2.564 2.464 NSCT-PCNN 10.423 5.305 4.459 4.482 4.417 4.667 NSST-PCNN 4.319 2.851 2.917 2.902 3.819 2.301 PCNN-IFS 5.429 1.226 3.194 2.631 2.197 2.903 NSST-FCM 6.061 4.102 5.269 3.462 4.556 4.321 NSST-PA-PCNN 3.589 3.452 3.454 4.839 3.543 3.455 Proposed Algorithm 4.063 2.091 2.658 2.286 2.981 2.2136 -
[1] WANG Z G, WEI W, SU B. Multi-sensor Image fusion algorithm based on multiresolution analysis[J]. International Journal of Online Engineering, 2018, 14(6): 44. doi: 10.3991/ijoe.v14i06.8697 [2] SHEN Yu, CHEN Xiaopeng, YUAN Yubin, et al. Infrared and visible image fusion based on significant matrix and neural network[J]. Laser & Optoelectronics Progress, 2020, 57(20): 56. [3] González José R, Damião Charbel, Moran Maira, et al. A computational study on the role of parameters for identification of thyroid nodules by infrared images (and comparison with real data) [J]. Sensors, 2021, 21(13): 4459. doi: 10.3390/s21134459 [4] XING Xiaoxue, LIU Cheng, LUO Cong, et al. Infrared and visible image fusion based on nonlinear enhancement and NSST decomposition[J]. EURASIP Journal on Wireless Communications and Networking, 2020(1): 125. [5] ZHANG Kang, HUANG Yongdong, YUAN Xia, et al. Infrared and visible image fusion based on intuitionistic fuzzy sets[J]. Infrared Physics and Technology, 2020(4): 85. [6] LIU Zhanwen, FENG Yan, CHEN Hang, et al. A fusion algorithm for infrared and visible based on guided filtering and phase congruency in NSST domain[J]. Optics and Lasers in Engineering, 2017, 97(6): 23. [7] 王丽丽, 刘辉, 郭肇禄. 图像融合在域内的算法研究进展[J]. 计算机技术与发展, 2021, 31(11): 8-14. https://www.cnki.com.cn/Article/CJFDTOTAL-WJFZ202111002.htmWANG Lili, LIU Hui, GUO Zhaolu. Research progress of image fusion algorithm in domain [J]. Computer Technology and Development, 2021, 31(11): 8-14. https://www.cnki.com.cn/Article/CJFDTOTAL-WJFZ202111002.htm [8] 徐慧娴, 黄坤超, 陈明举, 等. 基于深层次多分类生成对抗网络的红外与可见光图像融合方法[J/OL]. 电子测量技术, 2022, 3: 1-10.XU Huixian, HUANG Kunchao, CHEN Mingju, et al. Infrared and visible image fusion method based on deep multi classification to generate countermeasure network [J/OL]. Electronic Measurement Technology, 2022, 3: 1-10 [9] 许丽娜, 周丹丹, 高翾, 等. 一种结合GIHS、NSST和PCA的全色多光谱影像融合方法[J/OL]. 武汉大学学报: 信息科学版, 2022, 3: 1-17.XU Lina, ZHOU Dandan, GAO Zhen, et al. A panchromatic multispectral image fusion method combining GHIS, NSST and PCA [J/OL]. Journal of Wuhan University: Information Science Edition, 2022, 3: 1-17. [10] 李向阳, 曹宇彤, 陈笑, 等. 基于自适应NSST-PCNN的红外与可见光图像融合方法研究[J]. 长春理工大学学报: 自然科学版, 2021, 44(5): 12-18. doi: 10.3969/j.issn.1672-9870.2021.05.003LI Xiangyang, CAO Yutong, CHEN Xiao, et al. Research on infrared and visible image fusion method based on adaptive nsst-pcnn [J]. Journal of Changchun University of Technology: Natural Science Edition, 2021, 44(5): 12-18. doi: 10.3969/j.issn.1672-9870.2021.05.003 [11] SHEN Yu, YUAN Yubin, PENG Jing. Research on near infrared and color visible fusion based on PCNN in transform domain[J]. Spectroscopy and Spectral Analysis, 2021, 41(7): 65. [12] 刘帅, 王磊, 郝永平, 等. 基于NSCT与改进PCNN的红外与可见光图像融合方法研究[J]. 光电技术应用, 2021, 36(4): 60-65. https://www.cnki.com.cn/Article/CJFDTOTAL-GDYG202104017.htmLIU Shuai, WANG Lei, HAO Yongping, et al. Research on infrared and visible image fusion method based on NSCT and improved PCNN [J]. Optoelectronic Technology Application, 2021, 36(4): 60-65. https://www.cnki.com.cn/Article/CJFDTOTAL-GDYG202104017.htm [13] 戴文战, 王琪. 基于PCNN与IFS的可见光与红外图像融合方法[J]. 光电子·激光, 2020, 31(7): 738-744. https://www.cnki.com.cn/Article/CJFDTOTAL-GDZJ202007012.htmDAI Wenzhan, WANG Qi. Visible and infrared image fusion method based on PCNN and IFS [J]. Optoelectronics and Laser, 2020, 31(7): 738-744. https://www.cnki.com.cn/Article/CJFDTOTAL-GDZJ202007012.htm [14] WU Chenwen, MA Ning, JIANG Yufan. Weighted FCM clustering algorithm based on jeffrey divergence similarity measure[J]. Laser & Optoelectronics Progress, 2021, 58(8): 67 [15] 巩稼民, 刘爱萍, 张晨, 等. 基于FCM与ADSCM的红外与可见光图像融合[J]. 激光与光电子学进展, 2020, 57(20): 222-230. https://www.cnki.com.cn/Article/CJFDTOTAL-JGDJ202020025.htmGONG Jiamin, LIU Aiping, ZHANG Chen, et al. Infrared and visible image fusion based on FCM and adscm[J]. Progress in Laser and Optoelectronics, 2020, 57(20): 222-230. https://www.cnki.com.cn/Article/CJFDTOTAL-JGDJ202020025.htm [16] ZHANG Hongna, YAN Wei, ZHANG Chunyou, et al. Research on image fusion algorithm based on NSST frequency division and improved LSCN[J]. Mobile Networks and Applications, 2021, 26(5): 53. [17] GUO K, Labate D. Optimally sparse multidimensional representation using shearlets[J]. SIAM Journal on Mathematical Analysis, 2007, 39(1): 298-318. [18] Easley G, Labate D, LIM W Q. Sparse directional image representation using discrete shearlet transform[J]. Applied and Computation Harmonic Analysis, 2008, 25(1): 25-46. [19] YAN Peizhou, ZOU Jiancheng, LI Zhengzheng, et al. Infrared and visible image fusion based on NSST and RDN[J]. Intelligent Automation & Soft Computing, 2021, 28(1): 35. http://www.xueshufan.com/publication/3147917438 [20] TAN Wei, ZHANG Jiajia, XIANG Pei, et al. Infrared and visible image fusion via NSST and PCNN in multiscale morphological gradient domain[J]. Optics, Photonics And Digital Technologies For Imaging Applications Ⅵ, 2021, 11353: 54.