基于质量度量与颜色校正的多曝光图像融合算法
|
摘要
针对当前多曝光图像融合过程的图像质量属性选择不当,易导致颜色失真与丢失细节等问题,设计了一种质量度量耦合颜色校正的多曝光图像融合算法。首先,选择3种最显著的图像质量属性(对比度、饱和度和亮度)作为度量方式;然后,利用线性组合将得到的3种质量属性进行成加权组合,并采用幂函数来控制每个属性的影响,将较低的权重值赋予曝光不足和曝光过度的像素,消除具有视觉效果不佳的像素,从而有效保留正常的曝光像素及其明亮的颜色与细节;随后,将不同曝光图像加权组合的特征进行Laplacian金字塔分解,经过归一化权重映射后,对其进行不同系数的多分尺度融合,完成多曝光图像融合。为了避免颜色失真与细节丢失,采用基于局部饱和度的颜色校正方法来改善图像质量。实验数据表明,与当前多曝光图像融合方案相比,所提算法具有更高的融合视觉质量,可以更好地保持图像细节和校正曝光融合图像的颜色。
Aiming at the problems of color distortion and loss of detail caused by improper selection of image quality attributes in the current multi-exposure image fusion process,a multi-exposure image fusion scheme based on quality measurement coupled with color correction was designed. Firstly,three most prominent image quality attributes( contrast,saturation and brightness) were selected as measurement methods. Secondly,these three quality attributes were weighted by linear combination,and the power function was used to control the influence of each attribute. Low weight values are assigned to underexposed and overexposed pixels to eliminate pixels with poor visual effects,thus,effectively preserving exposure pixels,bright colors and details. Then,Laplacian pyramid decomposition is used to decompose the weighted combination features of different exposure images. After normalized weight mapping,multi-resolution fusion of different coefficients was performed to achieve multi-exposure image fusion. In addition,in order to avoid color distortion and detail loss,the post-processing steps of color correction based on local saturation are adopted to improve image quality. Experimental results show that the proposed algorithm has higher fusion visual quality than current multi-exposure image fusion scheme,and can better maintain image details and correct the color of the exposure fusion image.
Aiming at the problems of color distortion and loss of detail caused by improper selection of image quality attributes in the current multi-exposure image fusion process,a multi-exposure image fusion scheme based on quality measurement coupled with color correction was designed. Firstly,three most prominent image quality attributes( contrast,saturation and brightness) were selected as measurement methods. Secondly,these three quality attributes were weighted by linear combination,and the power function was used to control the influence of each attribute. Low weight values are assigned to underexposed and overexposed pixels to eliminate pixels with poor visual effects,thus,effectively preserving exposure pixels,bright colors and details. Then,Laplacian pyramid decomposition is used to decompose the weighted combination features of different exposure images. After normalized weight mapping,multi-resolution fusion of different coefficients was performed to achieve multi-exposure image fusion. In addition,in order to avoid color distortion and detail loss,the post-processing steps of color correction based on local saturation are adopted to improve image quality. Experimental results show that the proposed algorithm has higher fusion visual quality than current multi-exposure image fusion scheme,and can better maintain image details and correct the color of the exposure fusion image.
引文
[1]陈晔曜,蒋刚毅,邵华,等.高动态范围图像融合过程中的噪声抑制算法[J].光电工程,2018,45(7):74-84.CHEN Y Y, JIAN G Y, SHAO H, et al.Noisesuppression algorithmin the process ofhighdynamicrange image fusion[J]. Opto-Electronic Engineering,2018,45(7):74-84.
[2] CHIANG J C,HAO P H,CHEN Y S,et al. High-dynamic-range image generation and coding for multi-exposure multi-view images[J]. Circults Systems,andSignal Processing,2016,36(7):2786-2814.
[3] OH T H,LEE J Y,TAI Y W. Robust high dynamicrange imaging by rank minimization[J]. IEEETransactions on Pattern Analysis and MachineIntelligence,2015,37(6):1219-1232.
[4]李卫中,易本顺,邱康,等.细节保留的多曝光图像融合[J].光学精密工程,2016,24(9):2283-2292.LI W ZH,YI B SH,QIU K,et al. Detail preservingmulti-exposure image fusion[J]. Optics and PrecisionEngineering,2016,24(9):2283-2292.
[5]付争方,朱虹,薛杉,等.基于Sigmoid函数拟合的多曝光图像直接融合算法[J].仪器仪表学报,2015,36(10):2321-2329.FU ZH F,ZHU H, XUE SH, et al. Direct fusionalgorithm for multi-exposed images based on sigmoidfunction fitting[J]. Chinese Journal of ScientificInstrument,2015,36(10):2321-2329.
[6] MA K,LI H,YONG H. Robust multi-exposure imagefusion:A structural patch decomposition approach[J].IEEE Transactions on Image Processing,2017,26(5):2519-2532.
[7]张淑芳,刘孟娅,韩泽欣,等.基于Retinex增强的单幅LDR图像生成HDR图像方法[J].计算机辅助设计与图形学学报,2018,30(6):1015-1022.ZHANG SH F,LIU M Y,HAN Z X,et al. Generationmethod of high dynamic range image from a single lowdynamic range image based on retinex enhancement[J].Journal of Computer-Aided Design&ComputerGraphics,2018,30(6):1015-1022.
[8]芦碧波,李玉静,王玉琨,等.亮度分区自适应对数色调映射算法[J].计算机应用研究,2018,35(9):1-8.LU B B,LI Y J,WANG Y K,et al. Adaptive logarittunictone mapping arithmetic with luminance regionalization[J].Application Research of Computers,2018,35(9):1-8.
[9] KARA DUZOVIC-HADIABDIC K,TELALOVIC J H,MANTIUK R K. assessment of multi-exposure HDRimage de-ghosting methods[J]. Computers andGraphics,2017,63(1):1-17.
[10]侯春萍,马彤彤,岳广辉,等.基于高阶相位一致性的混合失真图像质量评价[J].激光与光电子学进展,2017,54(7):128-136.HOU CH P,MA T T,YUE G H,et al. Multiply-distorted image quality assessment based on high-orderphase congruency[J]. Laser&Optoelectronics Progress,2017,54(7):128-136.
[11]燕明,胡越黎,胡云生.保持局部对比度的动态范围压缩算法[J].上海大学学报(自然科学版),2017,23(5):682-689.YAN M, HU Y L, HU Y SH. Dynamic rangecompression with local contrast preservation[J]. Journalof Shanghai University(Natural Science Edition),2017,23(5):682-689.
[12] MERTENS T,KAUTZ J,REETH V F. Exposure fusion:a simple and practical alternative to high dynamic rangephotography[J]. Computer Graphics Forum, 2009,28(1):161-171.
[13]李成立,石俊生,徐领章,等.基于改进Retinex算法增强夜视融合图像的颜色传递[J].光电子·激光,2015,26(7):1381-1386.LI CH L,SHI J SH,XU L ZH,et al. Color transfer tonight vision fused image based on an improved Retineximage enhancement[J]. Journal of OptoelectronicsLaser,2015,26(7):1381-1386.
[14] PENG K,JUNG C,FANG Y. Perceptual multi-exposureimage fusion with overall image quality index and localsaturation[J]. Multimedia Systems,2017,23(2):239-250.
[15]卢鹏,林根巧,邹国良.基于信息熵和深度学习的无参考图像质量评价方法研究[J].计算机应用研究,2018,35(12):3895-3898.LU P,LIN G Q,ZHOU G L. Research on no-referenceimage quality assessment method based on informationentropy and deep learning[J]. Application Research ofComputers,2018,35(12):3895-3898.
[16] SURIN V A,TYSIN A N. Nonlinear filter model fordigital imaging of contrast images[J]. Optoelectronics,Instrumentation and Data Processing,2018,54(2):155-161.
[17]巩稼民,杨潇,杨萌,等.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(3):54-58.GONG J M,YANG X,YANG M,et al. An image fusionalgorithm based on region average gradient and regionenemy[J]. Journal of Xi'an University of Posts andTelecommunications,2016,21(3):54-58.
[18]李永生,桑庆兵.基于Contourlet变换的无参考立体图像质量评价[J].光学技术,2016,42(6):538-544.LI Y SH,SANG Q B. Quality evaluation of no referencestereo images based on Contourlet transform[J]. OpticalTechnology,2016,42(6):538-544.
[19]岳静静,李茂忠,陈骥,等.基于NSCT-PCNN的多聚焦红外图像融合[J].红外技术,2017,39(9):798-806.YUE J J,LI M ZH,CHEN J,et al. Multi-focus infraredimage fusion based on pulse coupled neural networks in anonsubsampled contour let transform domain[J].Infrared Technology,2017,39(9):798-806.
[20]张虎.图像质量评价主客观一致性的研究[D].武汉:武汉大学,2014:30-31.ZHANG H. Research on subjective and objectiveconsistency of image quality assessment[D]. Wuhan:Wuhan University,2014:30-31.
[21]耿卫江.结构相似度索引耦合最优稀疏表示的大规模损坏图像动态修复[J].科学技术与工程,2014,14(25):107-114.GENG W J. Large-scale damaged image dynamicrestoration algorithm based on structural similarity indexcoupled with optimal sparse representation[J]. Scienceand technology and Engineering,2014,14(25):107-114.
[2] CHIANG J C,HAO P H,CHEN Y S,et al. High-dynamic-range image generation and coding for multi-exposure multi-view images[J]. Circults Systems,andSignal Processing,2016,36(7):2786-2814.
[3] OH T H,LEE J Y,TAI Y W. Robust high dynamicrange imaging by rank minimization[J]. IEEETransactions on Pattern Analysis and MachineIntelligence,2015,37(6):1219-1232.
[4]李卫中,易本顺,邱康,等.细节保留的多曝光图像融合[J].光学精密工程,2016,24(9):2283-2292.LI W ZH,YI B SH,QIU K,et al. Detail preservingmulti-exposure image fusion[J]. Optics and PrecisionEngineering,2016,24(9):2283-2292.
[5]付争方,朱虹,薛杉,等.基于Sigmoid函数拟合的多曝光图像直接融合算法[J].仪器仪表学报,2015,36(10):2321-2329.FU ZH F,ZHU H, XUE SH, et al. Direct fusionalgorithm for multi-exposed images based on sigmoidfunction fitting[J]. Chinese Journal of ScientificInstrument,2015,36(10):2321-2329.
[6] MA K,LI H,YONG H. Robust multi-exposure imagefusion:A structural patch decomposition approach[J].IEEE Transactions on Image Processing,2017,26(5):2519-2532.
[7]张淑芳,刘孟娅,韩泽欣,等.基于Retinex增强的单幅LDR图像生成HDR图像方法[J].计算机辅助设计与图形学学报,2018,30(6):1015-1022.ZHANG SH F,LIU M Y,HAN Z X,et al. Generationmethod of high dynamic range image from a single lowdynamic range image based on retinex enhancement[J].Journal of Computer-Aided Design&ComputerGraphics,2018,30(6):1015-1022.
[8]芦碧波,李玉静,王玉琨,等.亮度分区自适应对数色调映射算法[J].计算机应用研究,2018,35(9):1-8.LU B B,LI Y J,WANG Y K,et al. Adaptive logarittunictone mapping arithmetic with luminance regionalization[J].Application Research of Computers,2018,35(9):1-8.
[9] KARA DUZOVIC-HADIABDIC K,TELALOVIC J H,MANTIUK R K. assessment of multi-exposure HDRimage de-ghosting methods[J]. Computers andGraphics,2017,63(1):1-17.
[10]侯春萍,马彤彤,岳广辉,等.基于高阶相位一致性的混合失真图像质量评价[J].激光与光电子学进展,2017,54(7):128-136.HOU CH P,MA T T,YUE G H,et al. Multiply-distorted image quality assessment based on high-orderphase congruency[J]. Laser&Optoelectronics Progress,2017,54(7):128-136.
[11]燕明,胡越黎,胡云生.保持局部对比度的动态范围压缩算法[J].上海大学学报(自然科学版),2017,23(5):682-689.YAN M, HU Y L, HU Y SH. Dynamic rangecompression with local contrast preservation[J]. Journalof Shanghai University(Natural Science Edition),2017,23(5):682-689.
[12] MERTENS T,KAUTZ J,REETH V F. Exposure fusion:a simple and practical alternative to high dynamic rangephotography[J]. Computer Graphics Forum, 2009,28(1):161-171.
[13]李成立,石俊生,徐领章,等.基于改进Retinex算法增强夜视融合图像的颜色传递[J].光电子·激光,2015,26(7):1381-1386.LI CH L,SHI J SH,XU L ZH,et al. Color transfer tonight vision fused image based on an improved Retineximage enhancement[J]. Journal of OptoelectronicsLaser,2015,26(7):1381-1386.
[14] PENG K,JUNG C,FANG Y. Perceptual multi-exposureimage fusion with overall image quality index and localsaturation[J]. Multimedia Systems,2017,23(2):239-250.
[15]卢鹏,林根巧,邹国良.基于信息熵和深度学习的无参考图像质量评价方法研究[J].计算机应用研究,2018,35(12):3895-3898.LU P,LIN G Q,ZHOU G L. Research on no-referenceimage quality assessment method based on informationentropy and deep learning[J]. Application Research ofComputers,2018,35(12):3895-3898.
[16] SURIN V A,TYSIN A N. Nonlinear filter model fordigital imaging of contrast images[J]. Optoelectronics,Instrumentation and Data Processing,2018,54(2):155-161.
[17]巩稼民,杨潇,杨萌,等.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(3):54-58.GONG J M,YANG X,YANG M,et al. An image fusionalgorithm based on region average gradient and regionenemy[J]. Journal of Xi'an University of Posts andTelecommunications,2016,21(3):54-58.
[18]李永生,桑庆兵.基于Contourlet变换的无参考立体图像质量评价[J].光学技术,2016,42(6):538-544.LI Y SH,SANG Q B. Quality evaluation of no referencestereo images based on Contourlet transform[J]. OpticalTechnology,2016,42(6):538-544.
[19]岳静静,李茂忠,陈骥,等.基于NSCT-PCNN的多聚焦红外图像融合[J].红外技术,2017,39(9):798-806.YUE J J,LI M ZH,CHEN J,et al. Multi-focus infraredimage fusion based on pulse coupled neural networks in anonsubsampled contour let transform domain[J].Infrared Technology,2017,39(9):798-806.
[20]张虎.图像质量评价主客观一致性的研究[D].武汉:武汉大学,2014:30-31.ZHANG H. Research on subjective and objectiveconsistency of image quality assessment[D]. Wuhan:Wuhan University,2014:30-31.
[21]耿卫江.结构相似度索引耦合最优稀疏表示的大规模损坏图像动态修复[J].科学技术与工程,2014,14(25):107-114.GENG W J. Large-scale damaged image dynamicrestoration algorithm based on structural similarity indexcoupled with optimal sparse representation[J]. Scienceand technology and Engineering,2014,14(25):107-114.