Exploring Illumination Removal of Face Image Via Logarithmic Gradient Penalty for Accurate Facial Expression Recognition

Xiaohe Li,Xingming Zhang
DOI: https://doi.org/10.1007/978-3-030-00214-5_127
2018-10-05
Abstract:Facial expression recognition system requires sufficient detailed image textures to distinguish emotions. However, the traditional Face Illumination Removal approach seem to break this condition providing illumination invariant inputs with heavy corrupted textures. To overcome this problem, we employed l0 gradient minimization approach to separate illumination variance and invariant components in the logarithmic face image. The experiment shows that, with carefully designed logarithmic gradient pruning mechanism, our method could boost the distinguish ability of any classifiers in any facial expression levels, especially the lower ones.
What problem does this paper attempt to address?