An Improved Algorithm for Segmentation of Motion Picture Image Sequences

Motion picture Special effects often require the segmentation of image sequences into foreground and background. The background and foreground may then be processed separately and and recombined, perhaps with additional elements in the scene. Alpha Estimation Systems can perform this segmentation where the background is natural (rather than an artificial background such as a blue screen) but they all require a user to create a hint image: an approximate segmentation of the image which is then refined automatically. A separate hint image is required for each frame of a sequence. This paper presents an algorithm which generates such hint images automatically, so that only a single input is required for an entire sequence. Results are presented which show our algorithm successfully generates hint images where an alternative approach fails. An adaptation to our previously published alpha estimation system is also shown which is capable of segmenting images even when the foreground is lit from behind.