Reconstruction Of The Surface Spectra Based On Digital Camera

The requirement to the reproduction of colors is much higher nowadays. The spectral reflectance, which is the best description of colors, is paid more attention to, since the colorimetric parameters as RGB, XYZ and CIELAB data can not express the target color accurately and truly. The spectrophotometer is a useful tool to measure the surface spectral reflectance, but it is not applicable for complex images or non-planar surfaces. On the other hand, digital cameras have been well developed, so they can adapt to color information capture under almost all kinds of conditions. Hereby, it is an important research direction in color science to reconstruct the spectral reflectance from RGB signals, which is now widely used in color printing and dyeing, high quality image acquisition and transfers, and conservation and restoration of arts.The aim of this dissertation is the spectral reconstruction based on signals of commercial digital camera. According to the various input signals for the spectral reconstruction models, two different techniques, the direct way (based on RGB signals from the camera) and the indirect way (based on XYZ signals resulted from colorimetric characterization of the camera), were studied, respectively, to recover the spectral reflectance, and their performances were analyzed and compared with each other.The device-independent tristimulus values XYZ, which were required in the indirect way of spectral reconstruction, were transformed from the device-dependent RGB signals by using the colorimetric characterization model of the camera. The polynomial regression model was adopted in this study to carry out the characterization. The experimental results implied that, the accuracy of the characterization model was related with the amount and type of the items in the polynomial. Moreover, different training and testing sample sets would lead to different results.As the key part of this study, the principal component analysis (PCA) method was employed to reconstruct the spectral reflectance, in which the principal coefficients were calculated by the polynomial model. It was indicated that the results of reconstruction were similar when the inputs were selected under individual illuminants, while the reconstruction performance was effectively improved by the introduction of multi-illuminant technique using the input signals under different illuminants together.The research of this dissertation showed that the spectral reconstruction models did have good accuracy for both the direct and the indirect ways, though the indirect reconstruction technique could give a little better result.Finally, the main conclusions this dissertation were summarized, and the perspectives of the future study were also forecasted.