Color-based detection robust to varying illumination spectrum

In color-based detection methods, varying illumination often causes problems, since an object may be perceived to have different colors under different lighting conditions. In the field of color constancy this is usually handled by estimating the illumination spectrum and accounting for its effect on the perceived color. In this paper a method for designing a robust classifier is presented, i.e., instead of estimating and adapting to different lighting conditions, the classifier is made wider to detect a colored object for a given range of lighting conditions. This strategy also naturally handles the case where different parts of an object are illuminated by different light sources at the same time. Only one set of training data per light source has to be collected, and then the detector can handle any combination of the light sources for a large range of illumination intensities.