Variational segmentation under variable illumination conditions

When illumination conditions vary significantly, the task of segmentation becomes more challenging since the existing ambiguities increase considerably. In this paper we introduce an illumination invariant framework that can be applied as a preprocessing step to obtain consistent segmentation when the lit conditions are substantially changed. We aim for an optimal luminance value that blends effectively the chromatic information with the initial luminance while the existing local features of the original lightness are enhanced. Practically, the local information is minimized by constraining the new lightness distribution with the chromatic contrast of the original image. We demonstrate the novel strategy for one of the main segmentation classes: variational PDE segmentation. Comprehensive qualitative and quantitative experiments prove the effectiveness and utility of the novel approach.