Unsupervised Learning of Motion for Distributed Video Coding
Introduction
We have developed a distributed video codec that uses Expectation Maximization to learn the motion only at the decoder. An overview of the codec is given in our journal paper [1]. More theory on how expectation maximization is used to learn motion/disparity in an unsupervised fashion can be found in earlier publications [2][3].
Package
The source code, written in ANSI C++, is being distributed as a Visual Studios package. Any Visual Studios version 2005 or later is compatible. The only external dependency is Matlab, whose functionality is called directly from the C++ code. Any Matlab version 7.0 or higher is compatible. The code has been thoroughly tested on Windows XP and Vista machines.
Visual Studios Project
LDPC Ladder File
Help Documentation
The code can be modified and used for research purposes. Your feedback is welcome.
References
[1] D. Varodayan, D. Chen, M. Flierl and B. Girod, "Wyner-Ziv coding of video with unsupervised motion vector learning", EURASIP Signal Processing: Image Communication. [Paper]
[2] D. Varodayan, Y.-C. Lin, A. Mavlankar, M. Flierl and B. Girod, "Wyner-Ziv coding of stereo images with unsupervised learning of disparity", Proc. Picture Coding Symposium, PCS 2007, Lisbon, Portugal, November 2007. [Paper]
[3] D. Varodayan, A. Mavlankar, M. Flierl and B. Girod, "Distributed grayscale stereo image coding with unsupervised learning of disparity", Proc. IEEE Data Compression Conference, DCC 2007, Snowbird, Utah, March 2007. [Paper]