100 likes | 316 Views
Image Registration. Department of Electrical and Computer Engineering. Narendhran Vijayakumar (Naren). 12/17/2007. Image Registration(1/2). Process of estimating the Geometric transformation function between two images. Rigid Transformation Rotational and Translational misalignment
E N D
Image Registration Department of Electrical and Computer Engineering Narendhran Vijayakumar (Naren) 12/17/2007
Image Registration(1/2) • Process of estimating the Geometric transformation function between two images. • Rigid Transformation • Rotational and Translational misalignment • Affine Transformation • Rotational, Translational, Shearing and Scaling
T2 FLAIR & diffusion weighted MRI (a) (b) (a) A T2 FLAIR diffusion weighted MR image of a patient with ROI marked. (b) Corresponding Diffusion weighted MR image of the patient.
Commercial Software • Philips Syntegra • Cross Correlation • Local Correlation • Mutual Information (MI) • BrainScan • Mutual Information
Registration Algorithms • Anatomical point landmark based methods • Labor intensive, Accuracy depends on user • Each slice has different structure • Surface-based registration • Limited by surface segmentation error • Voxel based registration • Cross correlation, Local correlation • Mutual information • Fourier Based Registration
Mutual information • Quantity that measures mutual dependence of two random variables • Maximum MI implies good registration • Search for (tx,ty,βz) where MI is maximum • Definitions
Entropy • Measures dispersion of probability density function. H(A,B) for misaligned images. Source: Pluim et al., “Mutual-information-based registration of medical images: a survey”, IEEE transactions on medical imaging, Vol 22, no 8, Aug 2003.
MI on Distributed system Algorithm • Step1: Master sends the Reference image I1 and floating image I2 to the workers. • Step2: Master then sends the range of search space to the workers. • Step 3: Workers receives the images I1 and I2, and find the global maxima (MI) for their given search space. • Step 4: Workers send the global maxima MI and the corresponding X and Y locations to the Master. • Step 5: Master Receives the MI and the corresponding X, Y values from all the workers. • Step 6: The maximum MI among the received values is the global maxima.
Conclusion • Up to 6 times speed up was achieved. • Speed up limitation • Saturates at when Np=5 • Amdahl’s Law • Fourier based registration • May be more fast