TexCut: A Study of Background Subtraction by Texture Comparison on Graph Cut

We propose a novel method for background subtraction named “TexCut,” which extracts object regions precisely even when change in lighting condition is rapid and spatially inhomogeneous. Human’s cast shadows and secondary reflections affect lighting condition at a human-object interaction. Texture-based approach is known as a robust approach for such affection: however it fails in regions with homogenious color, and with inhomogenious affection in the lighting condition. TexCut deal with homogenious region by using graph cut for propagating the difference in neighbor textured regions. For the second problem, we use spatial smoothness of the afferction under a planer light source. We experimentally evaluated our method in several scenes with different lighting environments.