Robust bilayer video segmentation by adaptive propagation of global shape and local appearance.

Authors: Soochahn Lee, Il Dong Yun, Sang Uk Lee

TL;DR: A robust video segmentation method that separates foreground from background by adaptively propagating both global shape priors and local appearance models across frames.

What this paper is about

Bilayer video segmentation – separating a video into foreground and background layers – is fundamental for video editing, compositing, and analysis. The challenge is maintaining accurate segmentation across frames when objects deform, lighting changes, or the background is cluttered.

Key idea

Flowchart of the proposed method.

The method combines global shape information with local appearance cues, propagating both adaptively from frame to frame. This dual propagation strategy makes the segmentation robust to appearance changes and shape deformations that would trip up methods relying on either cue alone.

Why it matters

Published in the Journal of Visual Communication and Image Representation, this work addresses a core problem in video processing with a principled approach that balances shape and appearance, improving reliability for practical video editing workflows.