Spectral Reflectance Reconstruction Of Multispectral Images

The color of nature is rich and colorful,and reproduction of color has become more and more important as people's understanding of color is deeper and deeper.As for the traditional color reproduction.Maybe we get the same color as the ISO color exists through the tradition color reproduction.The spectral reflectance as the property of the surface of the object is unique.In traditional,the measurement method of spectral reflectance is by spectrophotometer.However,the measurement range of spectrophotometer is small.And it can only measure on the plane.So we must find other ways to measure the spectral reflectance of the surface of the object.In recent years,it has been the focus of the study of the study to measure the spectral reflectance by the combination of CCD digital camera and filters bank.Due to the limitation of spectrophotometer,it's emphasis to reconstruct spectral reflectance by the digital camera in the filed of color science.In order to get more accurate spectral reflectance,it has been the research focus to reconstruct spectral reflectance by multispectral image system in recent years,namely which consists of three-channel digital camera and optical filter.In this paper,we reconstruct spectral reflectance by compound partial least squares(CPLS),and compare with the way of normal partial least squares(PLS).As a result,it's the better reconstruction precision than the way of normal partial squares method.For the multispectral image systems,multi spectral images with different number of channels can be obtained by different filters.But for number of filters of multispectral image systems.There is no uniform rule.And the influence of the number of filters on the accuracy of spectral reflectance reconstruction is unknown.So,in this paper,we use different number of filters combined with CCD camera to form different multispectral image system.And we use partial least squares method and the base vector method combined with neural network to reconstruct the spectral reflectance.Through the experimental results.We can know,with the increase of the number of filters,the higher the accuracy of spectral reflectance reconstruction for the PLS and neural network.