Together with other observational methods, visible and near infra-red polarimetry can help to understand the morphology of magnetic fields in the neighborhood of several star-forming regions. Inside molecular clouds, this morphology can be deduced with the help of submm polarimetry but rarely in the same regions. When both observational methods are used for the same region, there is a gap in the spatial scales to correctly compare the Galactic magnetic field with the magnetic field probed inside the clouds. This thesis proposes the necessary steps to make this type of multi-scale analysis and to better understand the role that can be played by magnetic fields in stellar formation regions. The GF 9 region is the first region analysed with this method. Then, a study of the morphology of the magnetic field located in filamentary molecular clouds OMC-2 and OMC-3 is presented, followed by a multi-scale analysis of the Orion A region, the molecular cloud complex in which these clouds are embedded. The results covering both regions can be summarized as follows. It is statistically shown that the large scale morphology of the field is poloidal in the GF 9 region, and probably toroidal in the Orion A complex. On the smaller spatial scale of the envelopes of the clouds, the magnetic fields appear to be aligned with the fields at their periphery. On the spatial scale of cores, the poloidal magnetic field located in the vicinity of GF 9 is apparently twisted and entrained by the rotation of the core and ambipolar diffusion does not seem to be effective at the present time. In Orion A, the morphology of the fields can hardly be probed in active sites of stellar formation in OMC-2, and is strongly constrained by the effects of gravity in OMC-1. There is no evidence for turbulence in all the observed regions. All in all, the multi-scale analyses suggest that independently of the evolutionary state or of the range in mass of the star-forming regions, the magnetic field morphology is significantly affected on spatial scales similar to those of cores, in the same way that molecular cloud properties remain self- similar down to the spatial scales similar to those of cores.