Colour‐appearance modeling using feedforward networks with Bayesian regularization method— Part I: Forward model

In this article, a method of predicting colour appearance (from colorimetric attributes to colour‐appearance attributes, i.e., forward model) using an artificial neural network is presented. The neural network model developed is a multilayer feedforward neural network model for predicting colour appearance (FNNCAM for short). The model was trained by LUTCHI colour‐appearance datasets. The Levenberg–Marquardt algorithm is incorporated into the back‐propagation procedure to accelerate the training of FNNCAM and the Bayesian regularization method is applied to the training of neural networks to improve generalization. The results of FNNCAM obtained are quite promising. © 2000 John Wiley & Sons, Inc. Col Res Appl, 25, 424–434, 2000     

Artificial neural networks for colour prediction in leather dyeing on the basis of a tristimulus system

Computer‐assisted colour prediction and quality control have become increasingly important to the dyeing process in many consumer goods manufacturing industries, including textile and leather. The most challenging aspect concerns dye recipe prediction for the production of the required shade on a given substrate. Computer recipe prediction based on the conventional and widely used Kubelka–Munk model often fails under a variety of conditions. In the present investigation, an attempt has been made to develop an artificial neural network model to predict colour in terms of tristimulus values (X, Y, Z) given the concentration of dyes. An artificial neural network model was trained with 300 pairs of known input vectors, i.e. dye concentrations, and output vectors, i.e. colour parameters, using a backpropagation algorithm. The artificial neural network topology consists of three neurons in the input layer to represent the concentration of dyes, three neurons in the output layer to represent the tristimulus values X, Y, and Z, and five neurons in the hidden layer with a log‐sigmoid transfer function. The artificial neural network results showed a good level of colour prediction during the training and testing phase. The results also indicate that the artificial neural network has the potential to give better predictive performance than the conventional Kubelka–Munk model.     

A novel approach to using neural networks to predict the colour of fibre blends

This work is concerned with the colour prediction of viscose fibre blends, comparing two conventional prediction models (the Stearns–Noechel model and the Friele model) and two neural network models. A total of 333 blended samples were prepared from eight primary colours, including two‐, three‐, and four‐colour mixtures. The performance of the prediction models was evaluated using 60 of the 333 blended samples. The other 273 samples were used to train the neural networks. It was found that the performance of both neural networks exceeded the performance of both conventional prediction models. When the neural networks were trained using the 273 training samples, the average CIELAB colour differences (between measured and predicted colour of blends) for the 60 samples in the test set were close to 1.0 for the neural network models. When the number of training samples was reduced to only 100, the performance of the neural networks degraded, but they still gave lower colour differences between measured and predicted colour than the conventional models. The first neural network was a conventional network similar to that which has been used by several other researchers; the second neural network was a novel application of a standard neural network where, rather than using a single network, a set of small neural networks was used, each of which predicted reflectance at a single wavelength. The single‐wavelength neural network was shown to be more robust than the conventional neural network when the number of training examples was small.     

CMC 2000 chromatic adaptation transform: CMCCAT2000

CMCCAT97 is a chromatic adaptation transform included in CIECAM97s, the CIE 1997 colour appearance model, for describing colour appearance under different viewing conditions and is recommended by the Colour Measurement Committee of the Society of Dyers and Colourists for predicting the degree of colour inconstancy of surface colours. Among the many transforms tested, this transform gave the most accurate predictions to a number of experimental data sets. However, the structure of CMCCAT97 is considered complicated and causes problems when applications require the use of its reverse mode. This article describes a simplified version of CMCCAT97— CMCCAT2000—which not only is significantly simpler and eliminates the problems of reversibility, but also gives a more accurate prediction to almost all experimental data sets than does the original transform. © 2002 John Wiley & Sons, Inc. Col Res Appl, 27, 49–58, 2002     

Neural network approach to a colorimetric value transform based on a large-scale spectral dataset

To improve the transform accuracy of colorimetric values for digital colour reproduction, a neural network approach based on a large-scale spectral dataset was proposed. The presented dataset exhibited a particularly wide colour gamut and was partitioned into 12 subsets according to dominant wavelength and excitation purity. In each subset, the non-linearity between colorimetric values under source and target illuminant–observer combinations was simulated by a backpropagation neural network. The colorimetric transform accuracy of this approach was compared with several existing methods. The experimental results indicated that the proposed approach significantly enhanced the transform precision for colorimetric values, especially for the colours located in highly saturated regions.     

A study of colour emotion and colour preference. Part II: Colour emotions for two‐colour combinations

Eleven colour‐emotion scales, warm–cool, heavy–light, modern–classical, clean–dirty, active–passive, hard–soft, harmonious–disharmonious, tense–relaxed, fresh–stale, masculine–feminine, and like–dislike, were investigated on 190 colour pairs with British and Chinese observers. Experimental results show that gender difference existed in masculine–feminine, whereas no significant cultural difference was found between British and Chinese observers. Three colour‐emotion factors were identified by the method of factor analysis and were labeled “colour activity,” “colour weight,” and “colour heat.” These factors were found similar to those extracted from the single colour emotions developed in Part I. This indicates a coherent framework of colour emotion factors for single colours and two‐colour combinations. An additivity relationship was found between single‐colour and colour‐combination emotions. This relationship predicts colour emotions for a colour pair by averaging the colour emotions of individual colours that generate the pair. However, it cannot be applied to colour preference prediction. By combining the additivity relationship with a single‐colour emotion model, such as those developed in Part I, a colour‐appearance‐based model was established for colour‐combination emotions. With this model one can predict colour emotions for a colour pair if colour‐appearance attributes of the component colours in that pair are known. © 2004 Wiley Periodicals, Inc. Col Res Appl, 29, 292–298, 2004; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/col.20024     

A comparative study of a CRT colorimetric prediction model by neural networks and the models by conventional method

In order to produce desired colors on CRT screens, much work has been done on the problem of the CRT colorimetric prediction. However, it would take great pains to overcome the troubles such as the constant channel chromaticity, the gun or channel independence, and the screen background effect, etc., with the conventional prediction methods such as PLCC and PLVC models, etc. To solve such problems, we propose a completely different CRT colorimetric prediction model by using a set of Artificial Neural Networks (ANN), where a set of back‐propagation (BP) neural networks is used to perform a nonlinear conversion between RGB values and XYZ values. By comparing some typical conventional CRT colorimetric prediction models with our neural‐networks‐based model theoretically, the article indicates that our new model can overcome the troubles faced by the conventional models, and by experiment the article shows that our new model can yield a satisfactory prediction result. © 1999 John Wiley & Sons, Inc. Col Res Appl, 24, 45–51, 1999     

基于混合遗传算法的人工神经网络模型及其对有机化合物熔点的预测

为了避免BP神经网络在训练过程中收敛于局部极小的缺陷,采用自适应交叉变异、最优保存的混合遗传算法对BP网络的权值和阈值进行优化,从而提出一种新的基于混合遗传算法的神经网络模型.该算法首先对一给定的网络结构,采用混合自适应交叉变异和最优保存策略,取各自的长处,用尽可能少的搜索代数找到问题的最优解,从而既防止算法陷入局部最优,又保证算法有较好的平均适应值和最佳的适应值个体.采用上述优化策略的人工神经网络可明显改善收敛的稳定性和收敛速度,并确保网络收敛于全局极小点.人工神经网络运用于物性数据的预测是一个具有潜力和有待开发的领域.运用该模型,根据有机化合物的分子量、临界密度、正常沸点和偶极矩,对其熔点进行预测.预测结果表明：提出的混合遗传算法神经网络优于其他算法神经网络,而且预测结果优于文献上已有的Joback方程和许氏方程的计算值.     

A study of colour emotion and colour preference. Part III: Colour preference modeling

In this study three colour preference models for single colours were developed. The first model was developed on the basis of the colour emotions, clean–dirty, tense–relaxed, and heavy–light. In this model colour preference was found affected most by the emotional feeling “clean.” The second model was developed on the basis of the three colour‐emotion factors identified in Part I, colour activity, colour weight, and colour heat. By combining this model with the colour‐science‐based formulae of these three factors, which have been developed in Part I, one can predict colour preference of a test colour from its colour‐appearance attributes. The third colour preference model was directly developed from colour‐appearance attributes. In this model colour preference is determined by the colour difference between a test colour and the reference colour (L*, a*, b*) = (50, ?8, 30). The above approaches to modeling single‐colour preference were also adopted in modeling colour preference for colour combinations. The results show that it was difficult to predict colour‐combination preference by colour emotions only. This study also clarifies the relationship between colour preference and colour harmony. The results show that although colour preference is strongly correlated with colour harmony, there are still colours of which the two scales disagree with each other. © 2004 Wiley Periodicals, Inc. Col Res Appl, 29, 381–389, 2004; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/col.20047     

Spectrophotometric colour matching algorithm for top-dyed mélange yarn,based on an artificial neural network

Colour, the first element of quality control of textile products, is a complex subject relating to physical optics, psychology, and the human visual system. Colour matching remains one of the major problems in the textile industry. Mélange yarn is a class of textile product with a specific colour appearance, which colour is mainly affected by colour matching of the dyed fibres and their ratio for spinning rather than by the dyeing process. The existing colour matching models for mélange yarn derived from specific types of fibre or specific spinning processes are restricted by the adopted conditions and parameters of the model, resulting in low universal applicability and low accuracy. In this paper, a spectrophotometric colour matching algorithm based on the back-propagation (BP) neural network and its processes were proposed. The weighted average spectrum was predicted by a BP neural network, followed by recipe prediction from the weighted average with constrained least squares. The results showed that the average colour difference of practical samples, based on the prediction of nine blind testing targets, was 0.79 CMC (2:1) units if more than two a priori training samples were used. This result indicated the capability and practicality of accurate prediction of colour matching for top-dyed mélange yarn by this novel method.     

Methods for designing characterisation targets for digital cameras

Characterisation targets usually include a set of physical coloured samples. A characterisation model can be derived between the colorimetric values (tristimulus values) and camera responses (RGB values) taken from an imaging device such as a digital camera capturing the colours in the target. The performance of such a model is highly dependent upon the number of colours and the colour region in the characterisation target. An ideal characterisation target should provide accurate model prediction without requiring too many samples. In this paper, a computational method is presented for colour selections to train a camera characterisation model based on a fourth‐order polynomial model including 35 terms. Compared with other available methods, the newly developed method performed better. It is proposed that this method be applied to generate generic targets in terms of colorimetric values. These targets should work reasonably well for a wide range of materials.     

Achieving expected depth of shade in reactive dye application using artificial neural network technique

Achieving the expected depth of shade in the production of dyed goods is a very important aspect. It requires the termination of the process at the right time in other words, correct duration of dyeing should be used. Prediction of this duration for the application of reactive HE dyes on cotton fabric using artificial neural network (ANN) is reported. The results obtained from the network gives an average training error of around 1% in the prediction of the time duration for achieving the correct depth of shade. The trained network gives the same average error % when tested with other reactive HE dyes even when the input parameters selected are beyond the range of inputs, which were used for training the network.     

A soft computing approach to model the structure–property relations of nonwoven fabrics

A soft computing approach to model the structure–property relations of nonwoven fabrics for filtration use is developed. Because the number of samples is very limited, the artificial neural network model to be established must be a small‐scale one. Consequently, this soft computing approach includes two stages. In the first stage, the structural parameters are selected by using a ranking method, to find the most relevant parameters as the input variables to fit the small‐scale artificial neural network model. The first part of this method takes the human knowledge on the nonwoven products into account. The second part uses a data sensitivity criterion based on a distance method that analyzes the measured data of nonwoven properties. In the second stage, the artificial neural network model of the structure–property relations of nonwoven fabrics is established. The results show that the artificial neural network model yields accurate prediction and a reasonably good artificial neural network model can be achieved with relatively few data points by integrated with the input variable selecting method developed in this research. The results also show that there is great potential for this research in the field of computer‐assisted design in nonwoven technology. © 2006 Wiley Periodicals, Inc. J Appl Polym Sci 103: 442–450, 2007     

Modeling color change after spinning process using feedforward neural networks

This article is concerned with the reflectance spectra prediction based on a neural network developed for yarn from the roving reflectance spectra. The neural network developed is a multilayer feed‐forward network. The first system is wavelength dependent, but its performance is not very satisfactory. The scaled conjugate gradient algorithm is incorporated into the backpropagation procedure to reduce the training phase. Once the wavelength independence of the transformation is established, a second system, whose performances agree with the experimental curves, is proposed. Finally, this system is completed by the introduction of a yarn parameter: the count. The results of this later model are quite promising. © 2002 Wiley Periodicals, Inc. Col Res Appl, 28, 50–58, 2003; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/col.     

航空发动机磨损趋势预测模型研究

文章讨论了神经网络的BP算法和遗传算法,提出用遗传算法来优化BP神经网络,应用遗传算法训练神经网络权重,实现网络结构的优化,用优化后的BP人工神经网络建立了航空发动机磨损故障趋势预测模型,利用发动机的光谱监测数据作为预测磨损趋势的特征参数,进行了模型的训练和预测试验,并将该模型预测结果与BP算法和多元线性回归法的预测结果进行了比较,证明了基于遗传算法的人工神经网络是航空发动机磨损故障趋势预测的一种理想方法。     

利用量子化学特征的模糊人工神经网络预测咪唑啉衍生物缓蚀效率

针对咪唑啉衍生物的量子化学特征参数与缓蚀效率存在复杂非线性关系，在利用多因素方差分析判断其相关性的基础上建立以最高占据轨道能量、最低未占据轨道能量、分子偶极矩、单点能、硬度、软度、亲核进攻指数、亲电进攻指数、电子转移参数以及咪唑环上非氢原子静电荷之和等量子化学特征参数为输入，以缓蚀效率为输出的模糊人工神经网络。结果表明，咪唑啉衍生物的量子化学特征参数及其缓蚀效率之间具有非常显著的相关性，据此所创建的Takagi-Sugeno型模糊人工神经网络预测模型采用10-30-1网络结构，通过Momentum优化算法对其进行反复训练直至其均方误差小于容许收敛误差限0.005，训练、测试阶段模型输出值与期望值近似呈线性关系，决定系数为0.9999，关联度较高，验证阶段该模型亦表现出良好的可靠性。因此利用量子化学特征的模糊人工神经网络预测模型能够准确预测不同咪唑啉系列衍生物的缓蚀效率。     

BP神经网络在合成镁钙砂中的应用

利用人工神经网络技术 ,建立BP网络模型 ,通过网络的学习训练 ,能够比较准确地预测影响合成镁钙砂粉化率的因素     

An example of facade colour design of mass housing

The colours and architectural characteristics of building facades are the major factors affecting the general appearance of cities. When cities are examined from various perspectives, first impressions are obtained from the geometrical forms and facade colours of buildings. The facade colour arrangements should reflect the features of the region and buildings. In this context various features of natural and artificial environments such as plant life, water elements, climate, and historical texture should be examined, and a facade colour arrangement should be designed according to the examination results. In addition, the other factors effective in determining the colour and style of a building, such as social‐cultural background of the society and traditional and natural building materials, should not be forgotten because in some regions traditional buildings with special construction styles, materials, and colours create a specific identity for the settlements and cities. The aims of this article are to elucidate the colour contrast, colour arrangement, and colour design stages of mass housing and to explain the colour design of Bizimkent Mass Housing, which was constructed in a new dwelling zone in Istanbul, Turkey, as an example of such an arrangement. © 2002 Wiley Periodicals, Inc. Col Res Appl, 27, 291–299, 2002; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/col.10068     

Prediction of Key State Variables using Support Vector Machines in Bioprocesses

Support vector machines (SVM) are applied to the prediction of key state variables in bioprocesses, such as product concentration and biomass concentration, which commonly play an important role in bioprocess monitoring and control. A so‐called rolling learning‐prediction procedure is used to deal with the time variant property of the process, and to establish the training database for the SVM predictor, which is characterized with the rolling update of the training database. As an example, product concentration in industrial penicillin production is predicted, and a comparison is also made with three different artificial neural network architectures (FBNN, RBFN, and RNN). The test results indicate that a prediction accuracy of 1–3 % can be obtained for 4–40 h ahead prediction using the SVM, which is better than the best of the three artificial neural networks (ANNs). Moreover, for noise‐added training highly noisy data or small‐sample learning, the SVM also clearly outperforms FBNN, RBFN, and RNN.