Research On Spectral Reflectance Reconstruction Algorithm For Color Reproductio

Multispectral imaging technology is a hot research topic in color reproduction and heritage conservation in recent years.Multispectral acquisition technology for color reproduction requires the acquisition of light-independent and device-independent spectral reflectance,so that the color information of objects can be reproduced realistically and objectively.In this paper,we discuss the spectral reflectance reconstruction algorithm for color reproduction,including the improved spectral reflectance reconstruction algorithm and the optimization of training sample selection,which achieves good color reproduction results.The traditional principal component-based spectral reconstruction algorithm is improved for the requirements of color accuracy and spectral accuracy for color reproduction,and an inverse variance-weighted regression spectral reflectance reconstruction algorithm is proposed.The spectral reflectance is reconstructed by introducing an inverse variance weighting model to adjust the factor weights,normalizing the spectral reflectance training sample set,calculating a new covariance matrix,and decomposing the singular values of the weighted spectral reflectance sample set.The experiments show that the CIE DE2000 color accuracy is improved by 3.3% and the spectral reconstruction accuracy is improved by 2%,and the results show that the method can achieve good color reproduction.A semi-supervised nearest-neighbor propagation clustering algorithm based on the entropy weight method is proposed for the selection of training samples with the influence of color correlation.Based on the nearest neighbor propagation algorithm to select the clustering center,the entropy value is calculated for the spectral reflectance sample set according to the entropy weight method,and the spectral reflectance with the largest weight in each class is selected as the most representative sample in each class to participate in the spectral reconstruction.The experiments show that the CIE DE2000 color accuracy is improved by 34% and the spectral reconstruction accuracy is improved by 16%,and the method has a certain improvement in spectral reconstruction accuracy compared with other training sample selection methods.The training sample selection method with improved K-mean clustering is proposed to address the deficiency of randomness in the selection of the initial values of training samples.The geometric center of the training sample set is used as the initial value of the cluster center,the probability density function of the spatial distribution of the samples is constructed based on the Gaussian function,and the Euclidean distance is used as the measure of other cluster centers.The similarity between the spectral reflectance samples in the training sample set is measured based on the intra-cluster squared difference,and the sample closest to the center in each cluster subset is used as the training sample.Experiments show that the CIE DE2000 color accuracy is improved by 39% and the spectral reconstruction accuracy is improved by 24%.The method improves the spectral reconstruction chromaticity and accuracy compared to other training sample selection methods.A fuzzy K-nearest neighbor based training sample selection method is proposed around the problem of minimizing the sum of squared global errors in training sample selection in spectral reconstruction.A fuzzy similarity matrix is established,each spectral reflectance vector is given a weight,and the optimization objective function is based on the distance minimization principle to compare the similarity of the spectral reflectance vectors in each class with the cluster center and to determine the most representative spectral reflectance.Experiments show that the CIE DE2000 color accuracy is improved by 42% and the spectral reconstruction accuracy is improved by 31%.The method results in a significant improvement in both the chromaticity and accuracy of the reconstructed spectral reflectance compared to other training sample selection methods.This paper systematically explores the study of a spectral reflectance reconstruction algorithm oriented to color reproduction,and reconstructs spectral reflectance by making corrections to the model.Experimental results show that the method proposed in this paper achieves good color reproduction and meets the needs of image reproduction in the printing industry.