sharppriorlevelsetmethodimagesegmentation创新.ppt

Level Set Based Shape Prior Segmentation The shape comparison terms defined as follows (T Chan, W Zhu:2005 ) Instead of that, the shape comparison term is defined as: New model can be written as: A Multiphase Dynamic Labeling Model for Variational Recognition-driven Image Segmentation Purely intensity-based segmentation (Chan vese) Global shape prior. Selective shape prior by dynamic labeling. (Daniel Cremers, Nir Sochen, and Christoph Schnorr:2006) A Similarity-Invariant Formulation In a realistic segmentation problem, one generally does not know pose and location of objects. What if there are several familiar objects in the scene? Extension to two known objects The terms associated with the two objects were normalized with respect to the variance of the respective template: The General Case: Multiphase Dynamic Labeling Introduce a vector-valued labeling function for n = 2, four regions are modeled by the indicator functions: In the general case of an n-dimensional labeling function The extension of the dynamic labeling approach to up to m regions can be cast into a cost functional of the form: Add the transformation parameters associated with each template. where Energy Minimization Evolution of the segmentation Evolution of the decision functions Pose optimization Shape distances for level sets Invariance by intrinsic alignment Shape distance is invariant to scale and translation. (D Cremers, SJ Osher, S Soatto :2006) Geodesic Active Contours with Combined Shape and Appearance Priors  Shape prior (Rami Ben-Ari,Dror Aiger:2008) where Appearance prior Recognition-Driven Two-Dimensional Competing Priors Toward Automatic and Accurate Building Detection Introduce a novel recognition-driven variational framework toward multiple building extraction from aerial and satellite images (Konstantinos Karantzalos, Nikos Paragios:2009) Discussion How to extract shape from the training examples How to define shape