In this report, we present a novel prior knowledge representation of shape variation using diffusion wavelets and applied for medical image segmentation. One of the major advantage of our approach is that it can reflect arbitrary and continuous interdependencies in the training data. In contrast to state-of-the-art methods our framework during the learning stage optimizes the coefficients as well as the number and the position of landmarks using geometric (reconstructed surface) constraints. Saliency is encoded in the model and segmentation is expressed through the extraction of the corresponding features in a new data-set. The resulting paradigm supports hierarchies both in the model and the search space, can encode complex geometric and photometric dependencies of the structure of interest, and can deal with arbitrary topologies. In another hand, our report deals with a different model search methodology where we apply an approach related to active feature models; the location of landmarks is updated iteratively, using local features, and the canonical correlation analysis. We report promising results on two challenging medical data sets, that illustrate the potential of our method.