The goal of this thesis is to propose algorithms for the non-rigid image registration and 3D reconstruction of deformable sufaces from monocular videos. Sheets of paper or faces are typical surfaces we want to deal with. These problems are intrinsically ill-posed. Prior information has to be used. Their degree of specificity to the surface at hand may vary. For example, a paper is a surface with everywhere vanishing Gaussian curvature. This property can not be used for a face. These issues are important for tasks such as video augmentation by texture editing or by adding virtual 3D objects. The first part of this thesis tackles non-rigid image registration. Previous work on warp-based deformation modeling and estimation methods are firstly described. We then present two contributions. The first one is an efficient registration algorithm using a feature-driven parametrization of the warp. This extends compositionnal algorithms to cope with non-rigid warps. Our second contribution is a specific framework for non-rigid image registration in spite of self-occlusions. The main idea is to force the warp to collapse at the self-occlusion boundary. The second part of this thesis is concerned with 3D reconstruction of deformable surfaces.We use the low-rank shape model which represents the 3D shape as a linear combination of unknown shape bases. We propose a new way of looking at the low-rank shape model. Instead of considering it as a whole, we assume a coarse-to-fine ordering of the deformation modes. This allows us to propose a coarse-to-fine reconstruction algorithm which easily incorporates a perspective camera model, automatically selects the number of modes, and drastically reduces the level of ambiguity of the original low-rank shape model. Finally, we study deformation capture of untextured surfaces from 3D data using boundary information.