对抗曝光变化的自适应加权光谱重建方法研究  

Adaptive Weighted Spectral Reconstruction Method Against Exposure Variation

作　　者：梁金星 辛磊 程靖尧 周景 罗航 LIANG Jin-xing;XIN Lei;CHENG Jing-yao;ZHOU Jing;LUO Hang(School of Computer Science and Artificial Intelligence,Wuhan Textile University,Wuhan 430200,China;Engineering Research Center of Hubei Province for Clothing Information,Wuhan 430200,China;Hubei Province Engineering Technical Center for Digitization and Virtual Reproduction of Color Information of Cultural Relics,Wuhan 430079,China)

机构地区：[1]武汉纺织大学计算机与人工智能学院,湖北武汉430200 [2]湖北省服装信息化工程技术研究中心,湖北武汉430200 [3]湖北省文物颜色信息数字化与虚拟再现工程技术研究中心,湖北武汉430079

出　　处：《光谱学与光谱分析》2023年第11期3330-3338,共9页Spectroscopy and Spectral Analysis

基　　金：湖北省自然科学基金项目(2022CFB537,2020CFB386);国家自然科学基金项目(62175191)资助。

摘　　要：光谱是颜色信息的指纹,也是表征物质物化属性的重要特征,对于颜色的高保真复制和物质的精准识别分析具有重要应用价值。基于光谱重建的多光谱成像技术利用成像系统拍摄物体表面数字图像,并通过光谱重建计算得到物体表面的多光谱图像,能够克服传统RGB图像对成像条件的依赖性,更加精细化地表征识别物体。该技术相对于价格昂贵的高光谱相机而言,能够有效提升多光谱图像的空间分辨率和获取效率,降低设备成本。然而,无论是基于机器学习还是基于深度学习的光谱重建方法,在光谱重建应用时均对图像的曝光变化敏感,即光谱重建方法在一种曝光水平下建立的光谱重建模型,无法直接在另一曝光水平下进行光谱重建应用,否则将导致重建光谱曲线的形状特征偏离真实光谱,制约着光谱重建面向光照强度易变和光照不均性场景的应用。针对现有光谱重建方法对图像曝光水平变化敏感的问题,论文提出了一种基于根多项扩展的自适应加权光谱重建方法,首先利用根多项式对样本rawRGB图像数据进行扩展,并利用伪逆法建立光谱重建模型,以解决模型对曝光水平变化的敏感性,然后在光谱不变特征空间构建自适应加权矩阵,以进一步提升光谱重建精度。研究以理论实验和具体实验样本为基础,首先对现有光谱重建方法对曝光变化的敏感性进行分析,然后将该方法与现有同类型光谱重建方法进行对比,最后探讨了加权策略对该方法的影响。实验结果表明,现有先进光谱重建方法均对曝光水平变化敏感,而该方法能有效对抗曝光水平变化,且光谱重建的光谱均方根误差(RMSE)和色差(ΔE^(*)_(ab))显著低于现有同类方法。此外,在光谱不变特征空间构建自适应加权矩阵,对于该方法提升光谱重建精度至关重要。研究成果对开放场景下的高精度光谱重建具有重要应用价值The surface spectral reflectance of the object is regarded as the fingerprint of its color,and at the same time,it is also an important feature to characterize the physical and chemical properties of substances.Multispectral imaging technology that is based on spectral reconstruction can overcome the dependence of RGB images on imaging conditions.Meanwhile,it can effectively improve the spatial resolution and acquisition efficiency of multispectral images and reduce equipment costs.Different from the principle of multispectral cameras,multispectral imaging based on spectral reconstruction first capture the digital images of the object using a digital imaging system,and then the corresponding multispectral images are reconstructed using spectral reconstruction methods.However,due to the mechanism of current spectral reconstruction methods,for both machine learning and deep learning methods,they are sensitive to exposure change of the image in practice.This means the spectral reconstruction model established at one exposure level cannot be directly used at another exposure level,or the curve shape of the reconstructed spectral reflectance will deviate from the ground truth.The sensitivity to exposure changes of current spectral reconstruction methods has limited their application in open environments with variable illumination intensity and inhomogeneity.To deal with the problems of current methods,an adaptive weighted spectral reconstruction method based on polynomial root expansion is proposed in this paper.In the proposed method,the raw RGB response of samples is firstly expanded by the root polynomial,and then the spectral reconstruction model is established by the pseudo-inverse algorithm.It will ensure the proposed method will be against the exposure changes.After that,an adaptive weighting matrix is constructed in the spectral invariant feature space to improve the spectral reconstruction accuracy further.The proposed method is verified and compared with the existing method through theoretical experiments an

关 键 词：光谱分析 光谱重建 多光谱图像 根多项式 光谱不变特征 

分 类 号：O432.3[机械工程—光学工程]