Chroma-Keying Based on Global Weighted Sampling and Laplacian Propagation

Abstract

Chroma-keying is an important technique for image/video background replacement, which is heavily used in film production, video game industry and news casting. In chroma-keying system, the foreground objects are shot in front of solid color background. In conventional chroma-keying systems, difference and clustering based algorithms are mainly used to separate foreground from background. However, regions cannot be reliably segmented if transparency or blurring exists, and if foreground and background region are not completely separable in color space. This paper proposes a new method to automatically remove the background color and to accurately extract the foreground objects along with their transparency properties, especially for tiny objects and large transparent area. First, a threshold based method in both HSV color space and spatial gradient space is used to roughly segment the foreground/background regions. The background color is then propagated from known background region to unknown region. Finally, the foreground color and the transparency factor are estimated based on weighted global sampling. The experimental results show that the foreground objects can be accurately extracted even if large transparent regions exist.