Identification and model validation have been performed for decades using global response of samples or structures. Nowadays, the development of full field measurement techniques offers the opportunity to take benefit of a much larger amount of experimental data. Existing strategies that can handle full field data to estimate constitutive model parameters are limited to two-dimensional mechanical models. In some cases, such as finite strain plasticity, two-dimensional modeling is not acceptable and three-dimensional effects must be accounted for. In this paper, we propose an identification strategy that mixes intimately numerical simulations with digital image correlation and that can overcome the common limitations of existing methods. It is shown to be noise robust, to have a low sensitivity to initial parameter guess and to mesh refinement. Furthermore, the proposed strategy allows to discriminate between constitutive models that give similar global responses.