基于视感知特征的多光谱高保真降维方法研究

发布时间：2018-05-27 03:27

本文选题：多光谱 + 视觉特征函数　；参考：《光谱学与光谱分析》2017年01期

【摘要】：为解决多光谱数据在降维压缩过程中的颜色精度保持问题,提出一种基于人眼视觉感知特征的多光谱数据高保真降维压缩方法(VPCM)。研究首先依据人眼视觉响应的非线性解析特征,成功构建了同时综合人眼光谱特征与色度特征的变换函数,并通过进一步构造的优化函数对其进行修正,以针对不同的样本集找到最佳变换方向,而后利用修正后的视觉特征变换函数对光谱样本集进行空间变换(Γ(S)=C),然后利用主成分分析方法对经视觉特征函数变换后样本集光谱数据进行降维压缩处理,并通过逆变换重构出样本集光谱数据(Γ-1(C)=^S),进行降维评价。实验选取四类具有典型代表性的数据集作为测试样本,分别以D50/2°条件下的CIELab色差和75组典型照明光源(钨丝灯、荧光灯和LED灯)下的平均同色异谱指数(MMI)作为色度主要评价指标,同时对比了Lab-PQR和2-XYZ两种较为先进的光谱降维算法。实验结果为VPCM方法的MMI值最小,其次是LabPQR,而2-XYZ的表现较差;VPCM方法在75组光源下对四组样本集的平均重构色差ΔEab也为最小,且最大样本平均色差及方差均要小于其他两种方法;VPCM方法的重构光谱精度介于Lab-PQR和2-XYZ之间,Lab-PQR的重构光谱精度最高。实验结果显示新方法色度压缩精度整体优于对比的两种方法,在变换参考条件下具有良好的色差稳定性,能够较好的应用于多光谱数据色度高保真压缩。
[Abstract]:In order to maintain the color accuracy of multispectral data in the process of dimensionality reduction, a high fidelity dimensionality reduction method for multispectral data based on human visual perception is proposed. Firstly, according to the nonlinear analytical features of human visual response, a transform function combining the spectral and chromatic features of human eyes is successfully constructed and modified by further optimization function. To find the best transformation direction for different sample sets, Then, the modified visual feature transform function is used to transform the spectral sample set (螕 ~ S), and then the spectral data of the sample set transformed by the visual feature function are reduced by the principal component analysis (PCA) method. The spectral data of the sample set (螕 -1) are reconstructed by inverse transformation, and the dimension reduction evaluation is carried out. Four kinds of typical data sets were selected as test samples. The CIELab chromatic aberration at D50 / 2 掳and 75 typical lighting sources (tungsten filament lamp) were used respectively. The average isochromatic heterospectral index (MMI) of fluorescent lamp and LED lamp is used as the main evaluation index of chromaticity. Two advanced spectral dimensionality reduction algorithms, Lab-PQR and 2-XYZ, are compared at the same time. The experimental results show that the MMI value of VPCM method is the smallest, the second is LabPQR, and the average reconstructed chromatic difference 螖 Eab of 2-XYZ method for four groups of samples under 75 groups of light sources is also the smallest. The average color difference and variance of the maximum sample are smaller than those of the other two methods. The reconstructed spectral accuracy of the VPCM method is between Lab-PQR and 2-XYZ. The reconstructed spectral accuracy of Lab-PQR is the highest. The experimental results show that the new method is superior to the two contrast methods in the accuracy of chromaticity compression, and it has good color difference stability under the condition of conversion reference, and can be applied to the high fidelity compression of multispectral data.
【作者单位】：武汉大学印刷与包装系;
【基金】：国家自然科学基金项目(61275172) 国家文物局项目(2013-YB-HT-034) 国家重点基础研究发展计划项目(2012CB725302)资助
【分类号】：TP391.41

【相似文献】