Shadow removal for light field images
Yan Zhou, Huiwen Guo, Guoyuan Liang, Xinyu Wu · 2014
In this paper, we present an automatic method to remove shadows in light field images. Taking into account the internal structure of the light field data, depth map of the captured scene is extracted to calculate the surface normal. Using nonlocal matching by combining chromaticity, normal and spatial location information in an anisotropic window, the shadow confidence of each pixel is established. For effectively utilizing the prior knowledge, MRF (Markov random fields) is introduced to obtain the shadow label of each pixel iteratively. Once obtained the shadow labels of pixels, inpainting with an energy minimize framework is used to remove shadows. The experimental results on real data demonstrate good performance of this algorithm.