Interactive segmentation of image sequences with near-static backgrounds
P. Hillman · 2005
Segmentation of image sequences is a major and time consuming part of digital special effects production. Where the sequence is too complex to segment using fully automatic techniques it is often segmented entirely by hand. This paper presents a fast semi–automatic technique which allows interactive segmentation of image sequences with near–static backgrounds. An operator initially creates a very approximate segmentation of the scenes. A foreground probability map is automatically generated by using mixture models which represent known foreground and background colour distributions. The operator then iteratively refines the initial segmentation by thresholding the generated probability maps until a suitably accurate segmentation has been achieved. Straightforward sequences can be segmented with very little human interaction: significant effort is required only on the most challenging sequences. Results are presented which demonstrate the use of our system on a variety of image sequences. 1