The Modulation Transfer Function (MTF) is one of the key indicators regarding Image Quality of Earth Observation systems. It characterizes the level of contrast that can be maintained by the optical system and is monitored during the whole life of the satellite. Due to the strong acquisition constraints as well as the gradually complexity of its estimation with future systems, it becomes necessary to increase the reactivity with a method free of acquisition constraints. In this paper, we present a model able to estimate the absolute MTF (or blur) level as well as its prediction uncertainty without any reference image or user parameter using deep learning techniques.