Friedrich-Alexander-Universität Erlangen
Lehrstuhl für Mustererkennung
Martensstraße 3
91058 Erlangen
The detection of skin regions in images is an important pre-processing step for many computer vision applications like face detection or face tracking. However, skin processing using color information can be a challenging task as in images the appearance of skin is affected by different factors such as illumination, surrounding and ethnicity.
The interaction of light with human skin is a very complex process as skin consists of several translucent layers. The prevalent chromophores in skin are different types of melanin and hemoglobin. Besides the specific absorbance characteristics of skin, the light is also highly scattered at small-scale structures like collagen or the chromophores themselves. These factors generate a very specific spectral reflectance of skin which varies mainly due to ethnic origin, but also due to individual characteristics such as age, body part, makeup etc.
The aim of this research is the analysis of the interaction of light with human skin from computer vision's perspective. Therein, the following two main parts have to be addressed:
Existing skin detection methods use suitable color space transformations and model the distributions of skin and non-skin pixels to perform a classification. However, the robustness against varying illumination conditions is still a major problem, as well as the reliable detection of different ethnicities. These aspects are addressed within this project. Only few skin detection approaches use color constancy techniques to compensate for illumination. However, the knowledge on the illuminant color can be useful for obtaining illuminant-specific skin color distributions. Furthermore, in other areas, like medicine or computer graphics, detailed physics-based models for the interaction of human skin and light exist. During this research a mapping from the complex physics-based spectral models to simpler computer vision models will be investigated.
As illumination compensation can be used to enhance skin segmentation algorithms, there is also the reverse possibility to use previously detected skin regions to estimate the illuminant color. The aim of this project is the enhancement of color constancy approaches by incorporating prior knowledge of the specific reflectance of skin, like the refractive indices.
Group members in these projects: 
Eva Eibenberger, Elli Angelopoulou