Related Experiment Videos
Shape modelling using Markov random field restoration of point correspondences
Rasmus R Paulsen1, Klaus B Hilger
1Oticon Research Centre, Eriksholm Kongevejen 243, DK-3070 Snekkersten, Denmark. rrp@imm.dtu.dk
Information Processing in Medical Imaging : Proceedings of the ... Conference
|September 4, 2004
Summary
This study introduces a new statistical shape model using Markov random field regularization for improved shape reconstruction and reduced variance. The method enhances correspondence finding in 3D data, like human ear canals.
Area of Science:
- * Computational anatomy
- * Medical imaging
- * Statistical modeling
Background:
- * Statistical shape models (SSMs) are crucial for analyzing anatomical variations.
- * Traditional SSMs can struggle with accurate correspondence and shape reconstruction.
- * Existing methods often lack robust regularization for complex shape fields.
Purpose of the Study:
- * To propose a novel method for building statistical point distribution models.
- * To improve the homogeneity and reconstruction capabilities of generative shape models.
- * To reduce the total variance within the point distribution model.
Main Methods:
- * Adaptation of Markov random field (MRF) regularization for the correspondence field.
- * Application of MRF regularization over a set of shapes.
- * Utilizing shape tangent space for identifying correlated semi-landmarks.
Main Results:
- * Development of a generative model producing highly homogeneous polygonized shapes.
- * Significant improvement in the reconstruction capability of training data.
- * Overall reduction in the total variance of the point distribution model.
- * Successful identification of highly correlated semi-landmarks.
Conclusions:
- * MRF regularization offers a powerful approach for enhancing statistical shape modeling.
- * The proposed method improves shape analysis and reconstruction accuracy.
- * Demonstrated effectiveness on 3D human ear canal data.