AMOS - An Active MPEG-4 Object Segmentation System
Di Zhong and Shih-Fu Chang
{dzhong,sfchang}@ee.columbia.edu
1. SYSTEM OVERVIEW
AMOS is an active object segmentation and tracking system for general video sources. It combines low level automatic region segmentation with an active
method for defining and tracking high-level semantic video objects.The system allows users to identify a semantic object (e.g., by MOUSE) in the starting frame of a video shot. The object is defined by a outline polygon
whose vertices and edges are roughly along the desired object boundary. Users can then start the object tracking process by specifying a few parameters (or using dafault ones). At any frame, users may stop the
tracking process to refine the object boundary, change the tracking parameters, and resume the tracking process. The tracking process contains two stages: an initial object segmentation stage where the user input is used to create a semantic object with underlying
homogeneous regions; and an object tracking stage where homogeneous regions and the object are tracked through successive frames.
The input video format can be ppm sequences or MPEG compressed format. The system output generates a binary-mask file (in the PGM format)
for the tracked object at each frame.
The AMOS system has the following main unique features:
- Tracking of generic video objects based on robust tracking of underlying regions
- Fusion of color, edge and motion in one integrated tracking process
- Region level motion estimation and compensation
- Tracking of both foreground and background objects
- Effective region aggregation and boundary alignment
- Active user inputs
- Platform independent software prototyping (in Java)
Click to Download AMOS Package
2. SYSTEM REQUIREMENT
The system is developed in Java language on Sun Solaris 2.5 and Windows NT 4.0. It requires following software packages.
- JDK 1.1.6+ or 1.2 (with JIT compiler)
- Java Media Framework (JMF) 1.1
In addition, the system uses the mpeg2decode from MPEG Software Simulation Group to decode frames from MPEG video sequences. Compiled decoders (i.e. executable
binary files) for Sun Solaris and WINNT are included in this distribution.
3. REFERENCES
Detailed explanation of the system and segmentation algorithms can be found
in reference papers.
- D. Zhong and S.-F. Chang, "An Integrated System for Content-Based Video Object Segmentation and Retrieval, part I: Semantic Video Object Segmentation and Tracking," IEEE Transactions on Circuits and Systems for Video Technology, 1998
- Di Zhong and Shih-Fu Chang, "AMOS: An Active System For MPEG-4 Video Object Segmentation", IEEE International Conference on Image Processing,
October 4-7, 1998, Chicago, Illinois, USA
- Di Zhong and Shih-Fu Chang, "Spatio-Temporal Video Search Using the Object Based Video Representation",IEEE International Conference on Image Processing, October 26-29, 1997, Santa Barbara, CA
Di Zhong and Shih-Fu Chang, "Video Object Model and Segmentation for Content Based Video Indexing",IEEE International Conference on Circuits and Systems, June, 1997, Hong Kong