S. Aouadi and L. Sarry (France)
Registration - 2D/3D, X-ray imaging, mutual information, industrial and orthopedics applications.
A new approach to recover rigid pose of 3D objects from digital or digitized radiographs is presented. A hybrid rigid registration that combines both intensity and geometric in formation is performed within a dedicated optimization framework. Validation was carried out on heterogeneous objects (X-ray attenuation obtained from computed tomog raphy) and homogeneous (uniform X-ray attenuation) ob jects for both simulated and real radiographs. Pose is re covered within 0.1 mm in translation and 0.1 in rotation for both screen-film and computed radiographs thanks to spe cial care brought to: i) Parzen distribution model used in the mutual information (MI) estimator (data pre-sphering in the bivariate case and bandwidth estimation in the uni variate case); ii) a quasi-global optimization scheme based on stochastic clustering (SC) and variance reduction with delete-d Jackknife ; iii) a non-linear model for the response of screen-film radiographs.
Important Links:
Go Back
