Artwork imaging from 370 to 1630 nm using a novel multispectral system based on light‐emitting diodes

The recent use of multispectral systems as a noncontact method for analysis of artworks has already shown promising results. This study explains the application of a novel portable multispectral system based on light‐emitting diodes (LEDs) for artwork imaging. This method provides spectral information in a spectral range from 370 to 1630 nm with a 25 cm × 25 cm field of view by using two different image sensors in synchrony with 23 bands of irradiation. The spectral information for each point is estimated and validated using the pseudo‐inverse and spline interpolation methods for spectral estimation and three different evaluation metrics. The results of the metrics obtained with both estimation methods show a general good performance of the system over the whole spectral range. The experiments also showed that the selection of the training set for the pseudo‐inverse estimation has a great influence in its performance, and thus, it defines whether or not the pseudo‐inverse outperforms the spline interpolation method. The system is applied in situ to the study of Catalan art masterpieces, and the results demonstrate the potential of a cost‐effective and versatile system using various off‐the‐shelf elements to reconstruct color information and to reveal features not previously identified. © 2014 Wiley Periodicals, Inc. Col Res Appl, 40, 398–407, 2015     

Color mixing and color separation of pigments with concentration prediction

In this study, we propose a color mixing and color separation method for opaque surface made of the pigments dispersed in filling materials. The method is based on Kubelka–Munk model. Eleven different pigments with seven different concentrations have been used as training sets. The amount of concentration of each pigment in the mixture is estimated from the training sets by using the least‐square pseudo‐inverse calculation. The result depends on the number and type of pigments selected for calculation. At most we can select all pigments. The combinations resulted with negative concentrations or unusual high concentrations are discarded from the list of candidate combination. The optimal pigment's set and its concentrations are estimated by minimizing the reflectance difference of given reflectance and predicted reflectance. © 2008 Wiley Periodicals, Inc. Col Res Appl, 33, 461–469, 2008     

Michromatic scope for enhancement of color difference

A michromatic (microscope plus chromatic) scope is a device that enhances the color discrimination between two spectral color datasets. Three spectral filters are required, instead of the conventional red, green, and blue filters, for the implementation of a michromatic camera. In this study, we describe two approaches to the design of these filters: in the first case, the design is based on the direct optimization of the filter characteristics (transmittance), whereas in the second case, the design is based on the nonnegative tensor factorization (NTF) of the spectral datasets. A michromatic camera can be implemented using these filters along with compatible postprocessing in‐camera firmware. Here, we performed experiments with two color datasets: one comprising skin and vein colors, and one comprising skin and cosmetics colors. These were further divided into a training set and a test set. The filters were defined using the training set, and the operation of the filters was tested and magnified using the test set. Our experiments demonstrated that the proposed approaches are suitable for color discrimination. For the first color dataset, the enhancement produced using the optimized filters was up to 252% of the original value, and the average color difference ΔE was increased from 2.82 to 9.93. NTF and preprocessing further enhanced the ΔE up to 21.84. For the second color dataset, NTF and postprocessing enhanced the ΔE from 4.33 to 29.19. The proposed discrimination enhancement could be physically implemented in a designated digital charge‐coupled device camera with proper filter installation and compatible postprocessing. © 2010 Wiley Periodicals, Inc. Col Res Appl, 2010     

Compression of spectral data using Box‐Cox transformation

The principal component analysis technique is used for the compression of different spectral databases including the reflectance spectra of nonfluorescent surfaces as well as the spiky spectra of the total radiance factors of fluorescent samples. Before extraction of principal directions, the Box‐Cox transformation technique is used in its original as well as modified version to improve the efficiency of employed compression technique by increasing the degree of normality in the datasets. The employed techniques are evaluated in terms of spectral dissimilarity between the reconstructed and the actual spectra and colorimetric differences by the value of CIELAB color differences of them under D65 and A illuminants and 1964 standard observer. The datasets departures from normal distribution are also investigated. The results confirm the effectiveness of the Box‐Cox modification technique for the reducing of spectral dimensions of samples. © 2012 Wiley Periodicals, Inc. Col Res Appl, 39, 136–142, 2014     

Optimal eigenvectors of spectral datasets: sequential selection from one set vs a collection from two sets

In this paper, the influence of spectral datasets and the method of selection of the corresponding feature vectors on the compression and reconstruction of data is scrutinised. To fulfil this aim, two different sets of reflectance data with the least spectral similarity are selected from different sets of spectral databases and the most optimal eigenvectors are chosen using different strategies. Six and 12 arrangements of eigenvectors obtained from different individual or combined databases are then used for the compression of reflectance spectra of learning sets, as well as those that have not been used in extraction of eigenvectors. The validity of the desired reduced subspaces is assessed by computing the spectral errors between the actual and the reconstructed spectra of samples of learning sets. Moreover, the efficiencies of designed compressed subspaces are evaluated through the numbers of out‐of‐range reconstructed spectra, as well as the spectral and colorimetric deviations between the actual and compressed‐reconstructed reflectance spectra of samples of datasets that were not employed in learning sequence. The results show that in the restricted subspaces, i.e. six‐dimensional subspace, the most effective results are achieved when the reduced subspace is created from a collection of two separate sets of eigenvectors of two different datasets with the maximum degree of dissimilarity, and the reduced spaces that have been made from six eigenvectors of individual datasets lead to higher errors.     

Estimating spectral reflectance from camera responses based on CIE XYZ tristimulus values under multi‐illuminants

A new spectral reflectance estimation method based on CIE XYZ values under multi‐illuminants was proposed to obtain multi‐spectral images accurately by using digital still cameras. CIE XYZ values under multi‐illuminants were initially predicted from raw RGB responses by using a polynomial model with local training samples. Then, spectral reflectance was constructed from the predicted CIE XYZ values via the pseudo‐inverse method. Experimental results indicated that the new spectral reflectance estimation method significantly outperformed the traditional colorimetric characterization method without requiring extra training samples or greatly increasing computational complexities. © 2016 Wiley Periodicals, Inc. Col Res Appl, 42, 68–77, 2017     

Paramerism and reliable parameric correction

In this article, the parameric decomposition method is considered as a technique for batch correction using three process primaries. It is proved that the parameric decomposition technique using optimal process primaries is a suitable method for reliable batch correction. Besides, the evaluation of parameric correction error for estimation of metamerism index is discussed. The creation of optimal spectral dataset to derive the appropriate process primaries is presented and the estimation of optimal statistical colorants for color reproduction of paint system is also developed. A set of pigments as primaries is used to generate three sets of batches using Kubelka–Munk theory; each set includes 10,000 paramer samples. Three sets of batches are generated to have different distributions around the match sample of target. The performance of process primaries derived from the Munsell dataset is compared with the optimal estimated statistical colorants for reliable parameric correction. To achieve greater success in estimation of appropriate metamerism index and make the estimated index more reliable, the statistical colorants used for parameric correction should be achieved from the spectral dataset generated by real primaries being used for matching of the target color. The outcome of parameric correction for pairs having miscellaneous mismatches is investigated to find out if it is sensible to just consider mismatch threshold for pairs being corrected. Finally, the conceptual subject of paramerism is developed to verify which spectral essence could be called as the paramer of an individual target color being matched. © 2010 Wiley Periodicals, Inc. Col Res Appl, 2010     

Dictionary‐based estimation of spectra for wide‐gamut color imaging

With the widespread use of commercialized wide‐gamut displays, the demand for wide‐gamut image content is increasing. To acquire wide‐gamut image content using camera systems, color information should be accurately reconstructed from recorded image signals for a wide range of colors. However, it is difficult to obtain color information accurately, especially for saturated colors, if conventional color cameras are used. Spectrum‐based color image reproduction can solve this problem; however, bulky spectral imaging systems are required for this purpose. To acquire spectral images more conveniently, a new spectral imaging scheme has been proposed that uses two types of data: high spatial‐resolution red, green, and blue (RGB) images and low spatial‐resolution spectral data measured from the same scene. Although this method estimates spectral images with high overall accuracy, the error becomes relatively large when multiple different colors, especially those with high saturation, are arranged in a small region. The main reason for this error is that the spectral data are utilized as low‐order spectral statistics of local spectra in this method. To solve this problem, in this study, a nonlinear estimation method based on sparse and redundant dictionaries was used for spectral image estimation—where the dictionary contains a number of spectra—without loss of information from the low spatial‐resolution spectral data. The estimated spectra are represented by a mixture of a few spectra included in the dictionary. Therefore, the respective feature of every spectrum is expected to be preserved in the estimation, and the color saturation is also preserved for any region. Experiments performed using the simulated data showed that the dictionary‐based estimation can be used to obtain saturated colors accurately, even when multiple colors are arranged in a small region. © 2011 Wiley Periodicals, Inc. Col Res Appl, 2013     

Spectral analysis of blacks

The spectral behavior of different black surfaces including papers and fabrics are investigated in this study. Several colored pigments are mixed with the blacks in different concentrations to prepare black surfaces with different shades while a series of black dyestuffs are applied on textile materials to increase the ranges of black objects. The principal component analysis technique is applied to determine the actual spectral size of the reflectance dataset. The technique simply extracts the principal directions of spectral data and organizes them in restricted spectral spaces. Three different spectral spaces, i.e., the reflectance spectra, the Kubelka‐Munk function of reflectance as well as the inverse of reflectance factor are selected to present the samples in the restricted spaces. Based on the results, it is found that, there are no significant differences between the employed spaces and far from the employed spectral domains, black surfaces could be adequately described in a three‐dimensional space. The three extracted statistical colorants are used for reconstruction of reflectance spectra of samples while the root mean square error percentage and the color difference values under the standard observing condition confirm the suitability of such virtual primaries. The work is extended to reconstruction of spectral data from colorimetric information and the adequacy of such three‐dimensional space is reconfirmed. © 2011 Wiley Periodicals, Inc. Col Res Appl, 2012     

Representation and estimation of spectral reflectances using projection on PCA and wavelet bases

In this article, we deal with the problem of spectral reflectance function representation and estimation in the context of multispectral imaging. Because the reconstruction of such functions is an inverse problem, slight variations in input data completely skew the expected results. Therefore, stabilizing the reconstruction process is necessary. To do this, we propose to use wavelets as basis functions, and we compare those with Fourier and PCA bases. We present the idea and compare these three methods, which belong to the class of linear models. The PCA method is training‐set dependent and confirms its robustness when applied to reflectance estimation of the training sets. Fourier and wavelet bases allow good generalization; an advantage of wavelets is that they avoid boundary artifacts. The results are evaluated with the commonly used goodness‐of‐fit coefficient (GFC), and the reliability of the use of wavelets is proved. © 2008 Wiley Periodicals, Inc. Col Res Appl, 33, 485–493, 2008     

Color‐difference formula performance for several datasets of small color differences based on visual uncertainty

Visual uncertainty, while reported, is not used routinely when evaluating color‐difference formula performance in comparison with visual data; rather, data are analyzed assuming no uncertainty; that is, repeating the experiment would result in the identical average results. Previously, Shen and Berns developed three methods to determine whether a color‐difference formula was well‐fitting, under‐fitting, or over‐fitting visual data when visual uncertainty was considered, the method dependent on how the uncertainty was reported and the colorimetric sampling of the color‐difference stimuli. The “nonellipsoid standard error method” was used in the current analyses. Three datasets were evaluated: BFD‐P, Leeds, and Witt. For the BFD‐P data, incorporating visual uncertainty led to the same performance results as the average results, that CIEDE2000 was an improvement over CIE94, which was an improvement over CIELAB. For the Witt data, incorporating visual uncertainty led to the same performance results as the average results, that CIEDE2000 and CIE94 had equivalent performance, both an improvement over CIELAB. However, both formulas under‐fitted the visual results; thus, neither formula was optimal. For the Leeds dataset, the visual uncertainty analysis did not support the improvement of CIEDE2000 over CIE94 that occurred when evaluating the average results. Both formulas well fit the visual data. These analyses also provided insight into the tradeoffs between the number of color‐difference pairs and the number of observations when fitting a local contour of equal perceived color difference: In particular, increasing the number of observations was more important than increasing the number of color‐difference pairs. Finally, average standard error could be used to approximate visual uncertainty defined using STRESS. © 2010 Wiley Periodicals, Inc. Col Res Appl, 2011     

Set theoretic estimation for problems in subtractive color

The use of linear algebra and set theoretic estimation for problems in color science and imaging is reviewed. Through a product‐space formalism, the powerful projections onto convex sets (POCS) algorithm is extended to subtractive color systems satisfying convex constraints in the density domain. Several convex sets are defined, which are useful in color science and imaging, and projections onto these sets are presented. The usefulness of the new methods is demonstrated by applying them to three practical problems: (1) model‐based scanner calibration, (2) design of color scanning filters that are color mixture curves, and (3) colorant formulation. © 2000 John Wiley & Sons, Inc. Col Res Appl, 25, 333–348, 2000     

Reconstruction of reflectance spectra using weighted principal component analysis

The weighted principal component analysis technique is employed for reconstruction of reflectance spectra of surface colors from the related tristimulus values. A dynamic eigenvector subspace based on applying certain weights to reflectance data of Munsell color chips has been formed for each particular sample and the color difference value between the target, and Munsell dataset is chosen as a criterion for determination of weighting factors. Implementation of this method enables one to increase the influence of samples which are closer to target on extracted principal eigenvectors and subsequently diminish the effect of those samples which benefit from higher amount of color difference. The performance of the suggested method is evaluated in spectral reflectance reconstruction of three different collections of colored samples by the use of the first three Munsell bases. The resulting spectra show considerable improvements in terms of root mean square error between the actual and reconstructed reflectance curves as well as CIELAB color difference under illuminant A in comparison to those obtained from the standard PCA method. © 2008 Wiley Periodicals, Inc. Col Res Appl, 33, 360–371, 2008     

Statistical methods for analyzing color difference distributions

Multivalued measurands, such as spectral reflectance and tristimulus values, are usually analyzed by reducing the data to a single‐valued parameter, such as color difference. The variations in sets of color differences are nonnormal distributed. This article compares five statistical methods to determine the 95% tolerance limit on nine data sets of color differences. Published 2010 Wiley Periodicals, Inc. Col Res Appl, 36, 160–168, 2011;     

基于导函数层序聚类的光谱成分预测方法研究

多光谱成像技术通过增加颜色通道的维数,成功的实现了基于光谱的颜色复制。然而,由于其颜色信息维数较高,此方法在提高色度精度的同时引入了较大的计算及存储压力。为此,最常用的方法就是通过对光谱数据进行分组,并利用每组光谱数据集中的主成分向量来对各个光谱曲线进行线性表示,从而实现数据的降维处理。提出了1种新的针对光谱数据导函数曲线的聚类分析方法,并利用伪逆算法进行光谱重建;本研究采用孟塞尔光泽色卡及无光泽色卡作为实验数据集,并将提出的导函数聚类分析法与现有的主成分分析法、聚类分析法以及色相角分类法相比较,实验结果证明其颜色预测精度在色度匹配及光谱匹配方面均优于现有方法。     

Optimal selection of representative samples for efficient digital camera-based spectra recovery

For digital camera-based spectra recovery, the spectral reflectance of the object being imaged always needs to be accurately recovered using training samples from available database. Considering the heavy workload when using all samples in database as training samples in practice, a new representative samples selection method is proposed for efficient digital camera-based spectra recovery based on single RGB image. The representative simulation system is firstly constructed through correlation analysis of spectra recovery results of different systems, and based on the representative simulation system, a few number of representative samples are selected from the database based on minimum of the defined simulate spectra recovery error. The effectiveness of the proposed method is evaluated and compared with existing method. As the results show, the proposed method outperforms the existing methods, and the robustness of the selected representative samples is consistent with the database in practical applications.     

Databases for spectral color science

In color science, spectral representation and analysis of colors have become a common approach to study color‐related problems, e.g., accurate industrial color measurement or analysis of color images. In developing algorithms for spectral color science, one often relies on existing databases of reflectance color spectra. Since a number of these databases are easily available, the same databases are commonly used by different research groups. During year 2003 the most popular one of our publicly available spectral reflectance databases was visited over 600 times. In the present article, we describe these color spectra databases and analyze their utility for spectral color science. However, the article does not take the complexity of fluorescent surfaces into account. The aim of this article is to set a solid ground for the comparisons of different methods in the spectral color science. The databases presented here include measured color spectra of natural and man‐made objects as well as spectra of some sets of standard colors. In addition to the commonly used data sets, some new data sets, including a set of standard calibrated colors and a set of natural colors, measured with 10 nm spectral resolution are introduced. © 2006 Wiley Periodicals, Inc. Col Res Appl, 31, 381–390, 2006; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/col.20244     

Is it better to combine predictions?

We have compared the accuracy of the individual protein secondarystructure prediction methods: PHD, DSC, NNSSP and Predator againstthe accuracy obtained by combing the predictions of the methods.A range of ways of combing predictions were tested: voting,biased voting, linear discrimination, neural networks and decisiontrees. The combined methods that involve `learning' (the non-votingmethods) were trained using a set of 496 non-homologous domains;this dataset was biased as some of the secondary structure predictionmethods had used them for training. We used two independenttest sets to compare predictions: the first consisted of 17non-homologous domains from CASP3 (Third Community Wide Experimenton the Critical Assessment of Techniques for Protein StructurePrediction); the second set consisted of 405 domains that wereselected in the same way as the training set, and were non-homologousto each other and the training set. On both test datasets themost accurate individual method was NNSSP, then PHD, DSC andthe least accurate was Predator; however, it was not possibleto conclusively show a significant difference between the individualmethods. Comparing the accuracy of the single methods with thatobtained by combing predictions it was found that it was betterto use a combination of predictions. On both test datasets itwas possible to obtain a ~3% improvement in accuracy by combingpredictions. In most cases the combined methods were statisticallysignificantly better (at P = 0.05 on the CASP3 test set, andP = 0.01 on the EBI test set). On the CASP3 test dataset therewas no significant difference in accuracy between any of thecombined method of prediction: on the EBI test dataset, lineardiscrimination and neural networks significantly outperformedvoting techniques. We conclude that it is better to combinepredictions.     

Prediction of bio-oil yield during pyrolysis of lignocellulosic biomass using machine learning algorithms

This work aims to implement and use machine learning algorithms to predict the yield of bio-oil during the pyrolysis of lignocellulosic biomass based on the physicochemical properties and composition of the biomass feed and pyrolysis conditions. The biomass pyrolysis process is influenced by different process parameters, such as pyrolysis temperature, heating rate, composition of biomass, and purge gas flow rate. The inter-relation between the yield of different pyrolysis products and process parameters can be well predicted by using different machine learning algorithms. In this study, different machine learning algorithms, namely, multi-linear regression, gradient boosting, random forest, and decision tree, have been trained on the dataset and the models are compared to identify the optimum method for the determination of bio-oil yield prediction model. Analysis of the results showed the gradient boosting method to possess a regression score of 0.97 and 0.89 for the training and testing sets with root-mean-squared error (RMSE) values of 1.19 and 2.39, respectively, and overcome the problem of overfitting. Therefore, the present study provides an approach to train a generalized machine learning model, which can be employed on large datasets while avoiding the error of overfitting.