This thesis presents the development of an automatic elastic registration method based on matching of vascular graphs extracted from both pre-operative and intra-operative images. The method can fuse accurate pre-operative information onto an organ undergoing small to large deformations during surgery, to compensate for the limited details provided by intra-operative imaging modalities and improve the visualization of tumor(s), vasculature and other important internal structures. Although methods dedicated to non-rigid graph matching exist, they are not efficient when noise, topology changes, and large intra-operative deformations are present. The first contribution presented is a biomechanical graph matching method (BGM) that builds on the work of Serradell et al. (2015). BGM combines the Gaussian Process Regression (GPR) matching with a biomechanical model of the organ, as a mean to discard matching hypotheses which would lead to non-plausible deformations (Garcia Guevara et al., 2018). However, BGM is not robust to noise, only matches limited size graphs and has a high computation time. The second contribution is the Adaptive Compliance Graph Matching (ACGM) method (Garcia Guevara et al., 2019), which allows to efficiently find the best graph matches with a novel compliance-based search and an adaptive rigid to soft approach. This reduces the computation time by predicting first the most plausible matching hypotheses. It also reduces the sensitivity on the search space parameters and improves the registration quality. The proposed registration methods are evaluated with realistic synthetic and real porcine datasets, showing that ACGM is compatible with intra-operative constraints.