In this paper, we address the problem of the total visual features loss during visual servoing. We present a new method allowing to reconstruct these features even if the image is completely unavailable. The proposed method has been developed for a 6 degree-of-freedom (DOF) calibrated camera and a static landmark of interest which can be characterized by point features. It relies on a predictor/corrector pair coupled with a depth estimation algorithm. Numerous simulation results are provided and show the relevance of the proposed approach.