In this study, the issue of reconstructing the gradients of noisy full-field data is addressed within the framework of solid mechanics. Two approaches are considered, a global one based on finite element approximation (FEA) and a local one based on diffuse approximation (DA). For both approaches, it is proposed to monitor locally the filtering effect in order to adapt the uncertainty to the local signal to noise ratio. Both approaches are applied to a case study which is commonly considered as difficult in solid mechanics (open-hole tensile test on a composite laminate). Both DA and FEA are successful for detecting local subsurface damage from the measured noisy displacement fields. Indications are also provided about the compared performances of DA and FEA. It is shown that DA is more robust, but the downside is that it is also more CPU time consuming.