| Presentation |
| Type of Publication |
| Automatic Video-Object Segmentation for MPEG-4 Coding and
Object-Behaviour Analysis |
| Title |
|
| Authors |
| Presentation,
DFG Rundgespräch "Information Retrieval", Dagstuhl, 11. March 2003 |
| Published in |
| Current video-processing systems still treat
video as rectangular images without further structure. However,
humans watching the video immediately recognize acting objects as
semantic units. This semantic object separation is currently not
reflected at the technical side, making it difficult to
manipulate the video at the object level. Enabling object based
manipulation will introduce many new possibilities for working
with videos like composing new scenes from pre-existing
video-objects, providing the possibility for user-interaction
with the scene, or classifying the video-objects for
content-based retrieval. Because of the vast amount of data in a
video, a prerequisite for object-based video-processing is
automatic segmentation of the raw input video into semantic
units, the video-objects.This presentation outlines a
segmentation algorithm which is based on camera-motion
compensation and background subtraction to detect foreground
objects. Camera-motion compensation is carried out using a
feature-based short-term predictor, combined with a dense
long-term predictor. A background mosaic is reconstructed using
temporal filtering to remove the foreground objects. These
foreground objects are subsequently extracted by subtracting the
background from motion-compensated input images and applying
regularization using Markov Random Fields. The obtained object
masks are further used to identify the objects from a database of
annotated object shapes. By providing a behavioral model of the
objects, automatic analysis of object-behaviour can be carried
out. |
| Abstract |
|
automatic segmentation
MPEG-4
behaviour analysis
|
| Keywords |
| [PDF]
[BIB] [XML] |
| Downloads & Bib-Entries |
