An effective approach to reconstruct the normalized radar cross section (NRCS) image from Global Navigation Satellite System-Reflectometry (GNSS-R) Delay-Doppler Map (DDM) is proposed. It is physically based on the truncated singular value decomposition (TSVD), properly extended to the 2-D case (2-D TSVD). The proposed approach is tested against simulated noisy DDMs, where both the additive and multiplicative noise are accounted for. The latter, generally known as speckle, is modeled using a Rice distribution. Experimental results show that the 2-D TSVD can be successfully exploited to reconstruct the NRCS field from DDM noisy measurements. Moreover, an analysis on the spatial resolution which characterizes the reconstructed domain is undertaken: it shows that generally a nonuniform spatial resolution is achieved while an area of the observed scene presents a almost uniform resolution that can be useful for remote sensing purposes.