بخشی از مقاله انگلیسی:
Skin color reproduction may be considered the most important problem in the color reproduction of color film and color television systems.1 With the recent progress in various imaging systems2–5 such as multimedia, computer graphic, and telemedicine systems, skin color becomes increasingly important for communication, image reproduction on hard copy and soft copy, medical diagnosis, cosmetic development, and so on. Human skin is a turbid medium with multilayered structure.6,7 Various pigments such as melanin and hemoglobin are contained in the medium. Slight changes in structure and pigment construction produce rich skin color variation.8 Therefore it is necessary to analyze skin color on the basis of structure and pigment construction in reproducing and discerning various skin colors. In this paper the spatial distributions of melanin and hemoglobin in skin are separated by independentcomponent analysis (ICA) of a skin color image. ICA is a technique that extracts the original signals from mixtures of many independent sources without a priori information on the sources and the process of the mixture. ICA has been applied to various problems such as array processing, communication, medical signal processing, and speech processing.9 In the field of color image processing, Inoue et al.10 proposed a technique to separate each pigment from compound color images. Their research is reviewed in Section 2 in this paper. However, they could not obtain any practical results, since they assumed linearity among the quantities of pigments and observed color signals in the intensity domain, and in the intensity domain this linearity generally will not hold in practical applications. We improve on their technique by processing color signals in the density domain and applying the technique to analyze the skin color image. Furthermore, we apply the result of the analysis to the separation and synthesis of a facial color image. In Section 2 we review the independent-component analysis proposed in Ref. 10 for application to color image separation. In Section 3 skin color is modeled on the basis of the two pigments melanin and hemoglobin in the optical density domain. The results of the independentcomponent analysis for skin color images are shown in Section 4. In Section 5, separated and synthesized facial color images are shown.
2. INDEPENDENT-COMPONENT ANALYSIS
ICA is a technique that extracts the original signals from mixtures of many independent sources without a priori information on the sources and the process of the mixture. To apply ICA to color image separation, Inoue et al.10 considered that the quantities of the pigments that construct the color are the original signals from independent sources, the observed color signals are mixtures, and the pure color signals of the pigments indicate the mixing process of the quantities.10 In this section we describe this technique as developed in Ref. 10. Simplifying the description, we assume that the medium is constructed by two pigments and that it is captured by an imaging system with two color channels. This simplification does not prevent generalization of the problem except when the number of pigments is larger than the number of channels. This situation is discussed below.