Two fields that used to be well separated are converging: 3D graphics and video coding. The emergence of new standards (such as, for example, MPEG-4) which provide to final users multimedia services makes this convergence easier. Thanks tothe concept of object coding, a scene can finally be reconstructed by the composition of various objects in a given order.This thesis deals with the general framework of visual objects composition. The kinds of objects that can be composed are also various: video sequences, still pictures, synthetic 3D objects, etc. However, the computational power requested in order to perform the composition is still prohibitive without any dedicated hardware and becomes critical in the context of mobile devices.Algorithmic and architectural reviews are performed in order to determine points that are shared between each field and to emphasize their differences. Next, three (interlinked) major axes of thought are explored: data representation, data accessesand processing organisation.The concrete case of a mobile device for labiophony allows the direct application of previous results. The labiophony is a smartphone application where the speaker face is reconstructed from a triangle mesh with texture mapping. A single architecture, able to equally process video sequences and triangle meshes, is defined. Finally, the architecture is synthesized and mapped into a prototyping platform. A comparison between this architecture and existing solutions can be performed.
