10h00 : Maxime Martineau
Learning a Graph Model with a Graph-Based Perceptron in Classification Context
Many tasks in computer vision and pattern recognition are formulated as graph matching problems. Despite the NP-hard nature of the problem, fast and accurate approximations have led to significant progress in a wide range of applications. Learning graph matching functions from observed data, however, still remains a challenging issue. This paper presents an effective scheme to parametrize a graph model for object matching in a classification context. For this, we propose a representation based on a parametrized model graph, and optimize it to increase a classification rate. Experimental evaluations on real datasets demonstrate the effectiveness (in terms of accuracy and speed) of our approach against graph classification with hand-crafted cost functions.
10h30 : Gaëtan Galisot
Anatomical Segmentation of Structures in MRI Brain images Using an Atlas-driven Set of Hidden Markov Random fields
In this work, a local atlas-driven set of Hidden Markov Random fields (HMRF) is combined with local classifiers in order to correct systematic errors of the HMRF. This schema helps in having a more efficient segmentation of anatomical structures in MRI Brain images. We propose to improve a wrapper method dedicated to the post-correction of misclassified voxels coming from classical segmentation methods in several ways to make it compatible with an incremental segmentation method. First, the probabilistic information coming from the HMRF instead of the symbolic class label is used as a feature for the corrective classifiers. Second, several local background classifiers are learned instead of a unique one. Third, a way to modulate the weights of these background classifiers is introduced in order to better handle the problem of regions adjacency. Qualitative and quantitative results on 3D brain images show an improvement of the segmentation quality (+2.2%) compared to the usual correction.