Texture And Multi-angle Color Assessment Of Samples With Metallic Coatings

Metallic coatings have been adopted in automotive industries recently due to its unique property with large change of appearance at different illuminating and viewing geometries.Since the portable multi-angle colorimeters only have a limited number of geometries,how to apply the measurements under these geometries for the predictions under other geometries is one of the important research interests.Meanwhile,because of the inefficiency and uncertainty of visual judgements for textures of metallic coatings,their modelling based on digital images has become rather necessary and urgent in the industry.The texture data from visual experiments and the images taken by digital cameras were employed to develop two texture models,which were further analyzed with correlation coefficients between the perceptual and predicted texture.On the other hand,the multi-angle color prediction models were established using the relationship between the changing color and the Gaussian distribution of metallic flakes' orientation angles with respect to the coating surface,and their performances were verified with the calculated color difference between the measured and predicted color.In order to match the texture predicted by models to the visual perception,the perceptual texture data of 12 observers on 128 metallic coatings was firstly acquired by the psychophysical method of grey scale,and the validity of the data was verified with the statistical parameters of STRESS and CV.Then,the captured texture images of metallic coatings by a Nikon D3X camera were applied for texture prediction after dark field correction,spatial uniformity correction,and the colorimetric characterization.This study employed the polynomial regression to fit the characterization model,and analyzed the effect of the different combinations of RGB values on characterization accuracy.This study also compared the characterization accuracy when 35 metallic coatings being employed for training the characterization model with that of ColorChecker Digital SG charts being adopted.The characterization model trained with 35 metallic coatings was finally used to transform the RGB signals of sample images into the tristimulus values.Meanwhile,the Laplacian Pyramid and Grey Level Difference method were respectively applied to extracting the textures from the sample images.The Fourier energy,calculated from the images by using Laplacian Pyramid,was fitted against the perceptual texture using linear regression.And the Entropy calculated with Grey Level Difference method,selected among other texture features like Contrast and Mean,was also fitted against the perceptual texture.The calculated correlation coefficients between the visual and predicted texture indicate their good prediction accuracies of the texture prediction model based on the Laplacian Pyramid and that of the Grey Level Difference method.In order to develop the multi-angle color prediction model,the spectral reflectance factors and the tristimulus values of 85 metallic coatings were firstly acquired by an X-Rite MA98 multi-angle spectrophotometer under 19 geometries.The color prediction model for metallic coatings based on the spectral reflectance factors was built with the assumption of the Gaussian distribution between the orientation angles of metallic flakes and the corresponding numbers,combined with the energy loss of refraction.The spectral model was later reconstructed into the color prediction model based on the tristimulus values by transforming the spectral reflectance factors into the tristimulus values.The discussions of the calculated CIEDE2000 and CAM02-SCD color differences indicate that the different illuminants have little influence on the models,and the color difference between the predictions by the two models is also little.Besides,both models could predict colors from the measurements under only 4,5 or 6 geometries,respectively.It was found that the average prediction accuracy from the measurements under specific 4 or 5 geometries can be as good as that with the measurements under 6 geometries,demonstrating the flexibility and fine prediction performance of the two developed models.Finally,on the basis of summarizing the research work of this dissertation,the perspectives of the future study are mentioned.