面向颜色再现的光谱降维方法研究

目的为了解决颜色信息光谱反射率的高维数给颜色再现过程带来的诸多问题。方法利用主流的光谱降维方法(主成分分析,Lab PQR,WSPCA,WSPCAplus),实现利于颜色再现的光谱信息筛选,从光谱精度、色度精度及变光源色差稳定性等3个方面评价这些降维方法。结果实验证明,主成分分析(PCA)法的光谱精度最高,光谱误差只有0.0076,但是在色度精度及变光源色差稳定性方面表现不好;Lab PQR法在光谱精度和色度精度间达到了很好的平衡,但是当变换光源时,其色差稳定性较差;WSPCA法去除了光源的影响,其变光源后色差变化不大,并且同时兼顾了色度精度,但以牺牲光谱精度为代价,WSPCAplus在其基础上将光谱精度提高为0.0081。结论研究结果对颜色再现过程中降维方法的选择具有一定的指导作用。

Spectral Dimension Reduction Methods for Color Reproduction

To solve the problems in color reproduction process caused by the high dimension number of the color spectral reflectance information. Several mainstream spectral dimension reduction methods including principal component analysis, LabPQR, WSPCA and WSPCAplus were used to realize the spectral information filtering for color reproduction. Then these dimension reduction methods were evaluated from the aspects of spectral accuracy, colorimetric accuracy and color difference stability in transformation of the light source. The experiment proved that the spectral accuracy of PCA was the highest, the spectrum error was only 0.0076, but its performance in colorimetric accuracy and color difference stability in transformation of the light source was not good. LabPQR achieved a good balance between spectral accuracy and colorimetric accuracy, but the color difference was not stable when the light source was changed. WSPCA eliminated the influence of light source to keep the stability of the color difference, and guaranteed the colorimetric accuracy at the same time, but it sacrificed the spectral accuracy. Based on WSPCA, WSPCAplus improved the spectral accuracy to 0.0081. The results have a certain referencing value for the selection of dimension reduction method in color reproduction process.