Uncalibrated Three Dimensional Reconstruction Of Vessels From Angiograms Based On Two Views

In current clinic practice, angiography technique is an unparallel visualization tool for the blood vascular structures. How to accomplish three dimensional reconstruction of vessels to provide more powerful diagnostic value is still an active research topic in the field of medical imaging. In this thesis, three dimensional reconstruction problems in angiography technique are discussed and a new uncalibrated three dimensional vessel reconstruction approach is developed.Multiscale Gabor filters are utilized to get the local maximum points on the response map in angiography images. The vessel branch centerline tracking is achieved by graph searching and the vessel branch correspondence is acquired based on the branch points. The corner points are detected according to the curvature formula and the virtual points are removed based on the polygon approximation principle. A reference gradient technique gray combined with gray correlation technique is used to match the feature points for the following estimation of the epipolar geometry.As a result of the instability of the estimation of the fundamental matrix with 8-points algorithms, errors are caused by assessing the essential matrix through the fundamental matrix with unknown camera intrinsic parameters. A weighted normalized linear approach is proposed to estimation the essential matrix directly. The essential matrix is estimated more accurately through strengthening the singularity constraint.The epipolar equation correlated to an arbitrary point on the vessel centerline is acquired based on the essential matrix. The epilolar constraint associated with disparity constraint is utilized to match the corresponding points in each vessel branch. The rotational matrix R and the translation vector t which represent the transform relation between two camera coordinate system are estimated according to the essential matrix. The three dimensional coordinates of each point on the vessel centerline is then computed. The reconstruction error is reduced by computing the average value of depth. The cross-section is then reconstructed according to the GC model.Experiments carried of real angiogram data sets demonstrate that the proposed algorithm can estimate the epipolar geometry accurately and satisfactory three dimensional vascular structures can be reconstructed with low computational cost.