Related Experiment Videos
A new & robust information theoretic measure and its application to image alignment.
1Department of Computer & Information Sciences & Engr, University of Florida, Gainesville, FL 32611, USA.
Information Processing in Medical Imaging : Proceedings of the ... Conference
|September 4, 2004
Summary
We introduce cumulative residual entropy (CRE), a new information measure for random variables. CRE offers advantages over Shannon entropy, particularly in image alignment tasks using cross-CRE (CCRE) for improved noise tolerance and convergence.
Area of Science:
- Information theory
- Probability and statistics
- Computer vision
Background:
- Shannon entropy is a standard measure of information but has limitations.
- Image alignment is a crucial task in computer vision.
- Existing methods like mutual information can be sensitive to noise and have limited convergence.
Purpose of the Study:
- To introduce a novel information measure, cumulative residual entropy (CRE).
- To define cross-CRE (CCRE) for assessing relationships between random variables.
- To apply CCRE to the image alignment problem, demonstrating its advantages over mutual information.
Main Methods:
- Developed CRE based on the cumulative distribution of a random variable.
- Defined CCRE as a measure of shared information between two random variables.
- Applied CCRE to solve image alignment for 3D rigid and affine transformations.
Main Results:
- CRE is more general and mathematically robust than Shannon entropy.
- CCRE demonstrates superior noise tolerance and a larger convergence range in image alignment.
- Experiments on synthetic and real data validate the effectiveness of CCRE.
Conclusions:
- CRE is a powerful new tool for information measurement.
- CCRE offers significant advantages for image alignment and potentially other applications.
- The proposed method provides a more robust and versatile approach to information quantification.