Image Matting via Local Tangent Space Alignment

Image matting refers to the problem of accurately extracting foreground objects in images and video. The most recent work [13] in natural image matting relies on the local smoothness assumptions on foreground and background colors on which a cost function is established. The closed-form solution has been derived based on certain degree of user inputs. In this paper, we present a framework of formulating new cost function from the manifold learning perspective based on the so-called Local Tangential Space Alignment algorithm [25] where the local smoothness assumptions have been replaced by implicit manifold structure defined in local color spaces. We illustrate our new algorithm using the standard benchmark images and very comparable results have been obtained.