基于压缩感知耦合梯度下降的红外-可见光图像自适应融合算法
|
摘要
设计了一种压缩感知耦合梯度下降的IR-VI图像自适应融合方案。引入S-函数对IR图像进行预处理,增强其对比度。利用非下采样Contourlet变换对IR与VI图像分解,分别得到低频与高频系数。对低频系数,利用自适应区域平均能量准则对其进行融合,以减少边缘模糊。对于高频部分,引入压缩感知进行稀疏采样,再采用绝对最大值选择与自适应高斯区域标准差的融合规则,通过高斯模糊隶属度建立的自适应控制融合过程,并利用基于梯度下降迭代算法来求解稀疏信号,形成高频融合系数。通过逆NSCT生成最终融合图像。实验表明,与当前流行的红外-可见光融合算法比较,所提算法具有更高的融合质量,输出图像的信息更丰富,边缘与纹理更为清晰。所提算法具有较高的融合质量,在红外、安防以及模式识别等领域具有一定的应用价值。
An adaptive fusion scheme of infrared-visible light images with compressed sensing coupling gradient descent was proposed.The S-function was used to preprocess the IR image,and its contrast was enhanced.Use the Nonsubsampled contourlet transform(NSCT)to decompose the IR and the visible images,and get the low frequency and high frequency coefficients respectively,respectively.Contourlet and NSCT can be used to get the low frequency and high frequency coefficients respectively.Then,for the low frequency part,the adaptive regional average energy criterion is used to fuse the low frequency coefficients to reduce the edge blur.For the high-frequency part,the fusion rule of absolute maximum standard selection and adaptive Gauss regional difference,adaptive fuzzy control membership can be used to establish the fusion process by Gauss,the introduction of compressed sensing sparse sampling,and is solved by sparse signal gradient descent based on iterative algorithm.The final fusion image is generated by the inverse NSCT reconstruction.Experiments show that compared with the current popular infrared visible light fusion algorithm,the image information obtained in this scheme is richer,the edges and texture are clearer,the contrast and spatial resolution are improved,and the system is more consistent with the human vision system,with high efficiency and strong robustness.This algorithm has high fusion quality which has certain application value in IR,security and pattern recognition.
An adaptive fusion scheme of infrared-visible light images with compressed sensing coupling gradient descent was proposed.The S-function was used to preprocess the IR image,and its contrast was enhanced.Use the Nonsubsampled contourlet transform(NSCT)to decompose the IR and the visible images,and get the low frequency and high frequency coefficients respectively,respectively.Contourlet and NSCT can be used to get the low frequency and high frequency coefficients respectively.Then,for the low frequency part,the adaptive regional average energy criterion is used to fuse the low frequency coefficients to reduce the edge blur.For the high-frequency part,the fusion rule of absolute maximum standard selection and adaptive Gauss regional difference,adaptive fuzzy control membership can be used to establish the fusion process by Gauss,the introduction of compressed sensing sparse sampling,and is solved by sparse signal gradient descent based on iterative algorithm.The final fusion image is generated by the inverse NSCT reconstruction.Experiments show that compared with the current popular infrared visible light fusion algorithm,the image information obtained in this scheme is richer,the edges and texture are clearer,the contrast and spatial resolution are improved,and the system is more consistent with the human vision system,with high efficiency and strong robustness.This algorithm has high fusion quality which has certain application value in IR,security and pattern recognition.
引文
[1]Yang Y,Tong S,Huang S,et al.Multifocus image fusion based on NSCT and focused area detection[J].Sens.J.IEEE,2015,15(5):2824-2838.
[2]Zhao C H,Guo Y T,Wang Y L.A fast fusion scheme for infrared and visible light images in NSCT domain[J].Infrared Physics and Technology,2015,72(9):266-275.
[3]Meng F J,Song M,Guo B L.Image fusion based on object region detection and non-subsampled contourlet transform[J].Computers and Electrical Engineering,2016,62(8):375-383.
[4]王昕,吉桐伯,刘富.结合目标提取和压缩感知的红外与可见光图像融合[J].光学精密工程,2016,24(07):1743-1753.Wang Xin,Ji Tongbo,Liu Fu.Infrared and visible image fusion based on target extraction and compressed sensing[J].Optical Precision Engineering,2016,24(07):1743-1753.
[5]Wang X C,Yao L J,Song R X.A new infrared and visible image fusion algorithm in NSCT domain[J].Infrared Physics&Technology,2017,61(10):420-431.
[6]Kong W W,Wang B H,Lei Y.Technique for infrared and visible image fusion based on non-subsampled shearlet transform and spiking cortical model[J].Infrared Physics and Technology,2015,71(3):87-98.
[7]闫利,向天烛.NSCT域内结合边缘特征和自适应PCNN的红外与可见光图像融合[J].电子学报,2016,44(04):761-766.Yan Li,Xiang Tianzhu.Infrared and visible image fusion based on edge feature and adaptive PCNN in NSCT domain[J].Electronic Journal,2016,44(04):761-766.
[8]Zhang J,Sohel F,Bennamoun M.NSCT-based fusion method for forward-looking sonar image mosaic[J].IET Radar,Sonar and Navigation,2017,11(10):1512-1522.
[9]Samane K,Faramarz H.Link state routing based on compressed sensing[J].Wireless Personal Communications,2018,99(1):253-271.
[10]徐彦.基于梯度下降的脉冲神经元在线学习方法[J].计算机工程,2015,41(12):150-155.Xu Yan.On-line learning method of impulsive neurons based on gradient descent[J].Computer Engineering,2015,41(12):150-155.
[11]许磊,崔光茫,郑晨浦.基于多尺度分解和显著性区域提取的可见光红外图像融合方法[J].激光与光电子学进展,2017,54(11):111-120.Xu Lei,Cui Guangmang,Zheng Chenpu.Visible infrared image fusion method based on multiscale decomposition and significance region extraction[J].Laser&Optoelectronics Progress,2017,54(11):111-120.
[12]Sabine B,Eckart Z,Patrick C.The spatial profile of mask-induced compression for perception and action[J].Vision Research,2015,110(1):128-141.
[13]张雨浓,肖争利,丁思彤.带后续迭代的双极S函数激励的WASD神经网络[J].中山大学学报:自然科学版,2016,55(04):1-10.Zhang Yunong,Xiao Zhengli,Ding Sitong.WASD neural network excited by bipolar S function with subsequent Iteration[J].Journal of Sun Yat-sen University:Natural Science Edition,2016,55(04):1-10.
[14]Mathur N,Glesk I,Buis A.Comparison of adaptive neurofuzzy inference system(ANFIS)and Gaussian processes for machine learning(GPML)algorithms for the prediction of skin temperature in lower limb prostheses[J].Med Eng Phys,2016,38(10):1083-1089.
[15]Magnusson D.The cure for the hard core:the evolution of planning doctrines and organizational unbundling in the Stockholm regional energy system[J].International Planning Studies,2018,23(1):65-80.
[16]林卉,梁亮,张连蓬.方向对比度和区域标准差相结合的图像融合[J].计算机工程与应用,2014,50(06):31-34.Lin Hui,Liang Liang,Zhang Lianpeng.Image fusion combining directional contrast and regional standard deviation[J].Computer Engineering and Application,2014,50(06):31-34.
[17]蒋伊琳,佟岐,张荣兵.自适应梯度下降观测矩阵优化算法[J].计算机应用研究,2017,34(07):1950-1952.Jiang Yilin,Teng Qi,Zhang Rongbing.Adaptive gradient descent observation matrix optimization algorithm[J].Computer Application Research,2017,34(07):1950-1952.
[18]Zhang X L,Li X F,Feng Y C.Image fusion with internal generative mechanism[J].Expert Systems with Applications,2015,42(5):2382-2391.
[19]Zhao Chunhui,Guo Yunting,Wang Yulei.A fast fusion scheme for infrared and visible light images in NSCT domain[J].Infrared Physics&Technology,2015,72(8):266-275.
[20]陈炳权,刘宏立.基于全变分Retinex及梯度域的雾天图像增强算法[J].通信学报,2014,35(6):139-147.Chen Bingquan,Liu Hongli.Fog image enhancement algorithm based on total variation Retinex and gradient domain[J].Journal of Communications,2014,35(6):139-147.
[21]巩稼民,杨潇,杨萌.基于区域平均梯度与区域能量的图像融合[J].西安邮电大学学报,2016,21(03):54-58.Gong Jiamin,Yang Xiao,Yang Meng.Image fusion based on regional average gradient and regional energy[J].Journal of Xi'an University of Posts and Telecommunications,2016,21(03):54-58.
[22]谭云兰,汤鹏杰,夏洁武.基于自适应引导滤波的全景图像增强算法研究[J].井冈山大学学报:自然科学版,2018,39(4):34-42.Tan Yunlan,Tang Pengjie,Xia Jiewu.Research on panoramic image enhancement algorithms based on adaptive guided filtering[J].Journal of Jinggangshan University:Natural Science Edition,2018,39(4):34-42.
[2]Zhao C H,Guo Y T,Wang Y L.A fast fusion scheme for infrared and visible light images in NSCT domain[J].Infrared Physics and Technology,2015,72(9):266-275.
[3]Meng F J,Song M,Guo B L.Image fusion based on object region detection and non-subsampled contourlet transform[J].Computers and Electrical Engineering,2016,62(8):375-383.
[4]王昕,吉桐伯,刘富.结合目标提取和压缩感知的红外与可见光图像融合[J].光学精密工程,2016,24(07):1743-1753.Wang Xin,Ji Tongbo,Liu Fu.Infrared and visible image fusion based on target extraction and compressed sensing[J].Optical Precision Engineering,2016,24(07):1743-1753.
[5]Wang X C,Yao L J,Song R X.A new infrared and visible image fusion algorithm in NSCT domain[J].Infrared Physics&Technology,2017,61(10):420-431.
[6]Kong W W,Wang B H,Lei Y.Technique for infrared and visible image fusion based on non-subsampled shearlet transform and spiking cortical model[J].Infrared Physics and Technology,2015,71(3):87-98.
[7]闫利,向天烛.NSCT域内结合边缘特征和自适应PCNN的红外与可见光图像融合[J].电子学报,2016,44(04):761-766.Yan Li,Xiang Tianzhu.Infrared and visible image fusion based on edge feature and adaptive PCNN in NSCT domain[J].Electronic Journal,2016,44(04):761-766.
[8]Zhang J,Sohel F,Bennamoun M.NSCT-based fusion method for forward-looking sonar image mosaic[J].IET Radar,Sonar and Navigation,2017,11(10):1512-1522.
[9]Samane K,Faramarz H.Link state routing based on compressed sensing[J].Wireless Personal Communications,2018,99(1):253-271.
[10]徐彦.基于梯度下降的脉冲神经元在线学习方法[J].计算机工程,2015,41(12):150-155.Xu Yan.On-line learning method of impulsive neurons based on gradient descent[J].Computer Engineering,2015,41(12):150-155.
[11]许磊,崔光茫,郑晨浦.基于多尺度分解和显著性区域提取的可见光红外图像融合方法[J].激光与光电子学进展,2017,54(11):111-120.Xu Lei,Cui Guangmang,Zheng Chenpu.Visible infrared image fusion method based on multiscale decomposition and significance region extraction[J].Laser&Optoelectronics Progress,2017,54(11):111-120.
[12]Sabine B,Eckart Z,Patrick C.The spatial profile of mask-induced compression for perception and action[J].Vision Research,2015,110(1):128-141.
[13]张雨浓,肖争利,丁思彤.带后续迭代的双极S函数激励的WASD神经网络[J].中山大学学报:自然科学版,2016,55(04):1-10.Zhang Yunong,Xiao Zhengli,Ding Sitong.WASD neural network excited by bipolar S function with subsequent Iteration[J].Journal of Sun Yat-sen University:Natural Science Edition,2016,55(04):1-10.
[14]Mathur N,Glesk I,Buis A.Comparison of adaptive neurofuzzy inference system(ANFIS)and Gaussian processes for machine learning(GPML)algorithms for the prediction of skin temperature in lower limb prostheses[J].Med Eng Phys,2016,38(10):1083-1089.
[15]Magnusson D.The cure for the hard core:the evolution of planning doctrines and organizational unbundling in the Stockholm regional energy system[J].International Planning Studies,2018,23(1):65-80.
[16]林卉,梁亮,张连蓬.方向对比度和区域标准差相结合的图像融合[J].计算机工程与应用,2014,50(06):31-34.Lin Hui,Liang Liang,Zhang Lianpeng.Image fusion combining directional contrast and regional standard deviation[J].Computer Engineering and Application,2014,50(06):31-34.
[17]蒋伊琳,佟岐,张荣兵.自适应梯度下降观测矩阵优化算法[J].计算机应用研究,2017,34(07):1950-1952.Jiang Yilin,Teng Qi,Zhang Rongbing.Adaptive gradient descent observation matrix optimization algorithm[J].Computer Application Research,2017,34(07):1950-1952.
[18]Zhang X L,Li X F,Feng Y C.Image fusion with internal generative mechanism[J].Expert Systems with Applications,2015,42(5):2382-2391.
[19]Zhao Chunhui,Guo Yunting,Wang Yulei.A fast fusion scheme for infrared and visible light images in NSCT domain[J].Infrared Physics&Technology,2015,72(8):266-275.
[20]陈炳权,刘宏立.基于全变分Retinex及梯度域的雾天图像增强算法[J].通信学报,2014,35(6):139-147.Chen Bingquan,Liu Hongli.Fog image enhancement algorithm based on total variation Retinex and gradient domain[J].Journal of Communications,2014,35(6):139-147.
[21]巩稼民,杨潇,杨萌.基于区域平均梯度与区域能量的图像融合[J].西安邮电大学学报,2016,21(03):54-58.Gong Jiamin,Yang Xiao,Yang Meng.Image fusion based on regional average gradient and regional energy[J].Journal of Xi'an University of Posts and Telecommunications,2016,21(03):54-58.
[22]谭云兰,汤鹏杰,夏洁武.基于自适应引导滤波的全景图像增强算法研究[J].井冈山大学学报:自然科学版,2018,39(4):34-42.Tan Yunlan,Tang Pengjie,Xia Jiewu.Research on panoramic image enhancement algorithms based on adaptive guided filtering[J].Journal of Jinggangshan University:Natural Science Edition,2018,39(4):34-42.