Efficient and Robust Inverse Lighting of a Single Face Image Using Compressive Sensing

摘要: In this paper, we show that the recent theory of Compressive Sensing (CS) can successfully be applied to solve a model-based inverse lighting problem for single face images, even in harsh with multiple light sources, including cast shadows and specularities. It has been shown an illumination cone used perform realistic lighting. work, images are synthetically generated using directional lights reflectance faces. Thereby, model is achieved by fitting 3D Morphable Model input image. We apply CS find sparsest setup from few random measurements RGB images. The proposed method significantly reduces dimensionality through stochastic sampling greedy algorithm sparse support estimation, yielding low runtimes. search designed handle non-negativity sources joint-support selection. reaches quality estimation equal previous while dramatically reducing number active sources. Thorough experimental evaluation shows stable recovery achievable compression rates up 99%. exhibits outstanding robustness additive noise