Applications of Color Image Processing Based on Spectral Information

Color is one of the most important points of information about an object. The tristimulus values of CIE-XYZ and RGB values obtained by a CCD camera or scanner, however, are dependent on the characteristics of the imaging devices and illuminant conditions. Therefore we have proposed the use of the spectral reflectance of the object for color image processing, transmission and reproduction. From principal component analysis of gastric mucous membrane and skin color, we have shown that those reflectance spectra can be estimated respectively from the R, G, B signals of electronic endoscopes and a CCD camera. On the other hand, we have also shown that it is necessary to use five channels to estimate the reflectance spectra of oil paintings. We have developed a 5-band CCD camera with 3120H x 2160V pixels and 14 bits for recording the reflection spectra of the paintings and artifacts. The estimated reflection spectra were applied to improve color reproduction and image processing. In this paper, our recent research on the analysis and estimation of spectral reflectance, and some applications are introduced and presented. The first one is measurement of skin absolute spectral-reflectance image and the application to component analysis. Mapping the oxygen saturation of blood and melanin in the skin is expected to provide useful information for skin diagnosis. We have already proposed a technique for estimating the map of skin components from a spectral reflectance image obtained by multiband images based on optical scattering analysis. The spectral reflectance image is a result of dividing the spectral image of the object by the spectral image of an absolute white reference plane. However, the shape of skin is not 2-dimensional. The required absolute spectral reflectance is not obtained in the wide region of the skin by the conventional method. In this research, the absolute spectral reflectance and normal vector of the skin surface are obtained using a photometric stereo technique. In the photometric stereo technique, more than three illuminants are used, and an image is taken by each illuminant. The pixel values in the obtained multiple images are used to calculate the absolute spectral reflectance and normal vector of the surface for the corresponding pixel. An index finger where the second joint was bound by a string was captured and analyzed using the proposed technique. The results show that the components of 15 A ppations of C olor Im ge P rossing B ased on Spctral Iform ation pigmentation are extracted in the wide range of the finger. The second one is an estimation system for transmittance spectra and its application to an image database system. The transmittance spectra of fungi were estimated from multiband optical micrographs. An optical microscope modified with a singlechip CCD camera with several band filters was used for image acquisition. The Wiener method was applied to estimate the transmittance spectra of two species which belong to one genus of fungi. The Wiener estimation operator was calculated using the transmittance spectra of color samples and the corresponding camera responses. The estimated transmittance spectra were used for reproduction of fungal optical micrographs on three light sources, then a prototype of the illustrated fungi database was developed using the reproduced micrographs. The last one is measurement and analysis of reflectance spectra in the rectal mucous membrane. Using an endoscopic spectrophotometer, 71 reflectance spectra of rectal mucous membrane were measured from 16 patients. The measured reflectance spectra were analyzed by principal component analysis. The analysis shows that the reflectance spectra can be adequately represented by three principal components. On the basis of this experimental result, the reflectance spectra of each pixel in the rectal mucous membrane images taken by electronic endoscopes were estimated from RGB signals using three eigenvectors calculated by principal component analysis. The Wiener estimation method was also introduced to estimate the reflection spectra. Color reproduction of electronic endoscopes with different illuminants was estimated and analyzed by computer simulation. 16 A pp lic at io ns o f C ol or I m ag e P ro ce ss in g B as ed o n Sp ec tr al I nf or m at io n A. An Outline of Color Image Processing Based on Spectral Information