The horizontal to vertical spectral ratio (HVSR) of seismic ambient noise has been proven to be a fast and efficient method for characterizing the 1-D resonance frequency of the local subsurface in a practical framework. Over the last decades, theories have been developed in order to extend the exploitation of HVSR beside the frequency of its first peak, notably the diffuse field assumption (DFA) which links the HVSR to the Green's function of the local medium assuming the diffuseness of the seismic ambient noise wavefield. However, the underlying assumption of the seismic ambient noise being a diffuse, equipartitioned field may not be satisfied under certain circumstances. In order to exploit the contribution of scattering in forging diffuse wave fields, we leverage the advantages of coda waves and present a novel procedure for computing the HVSR, using the coda part of ambient noise correlations. We applied this technique to data gathered at the plio-quaternary sedimentary basin of Argostoli, Greece. Results on this data set show the potential of the method to improve the temporal stability of the HVSR measurements compared to the classical computation, and the fit with the theoretical HVSR curve derived from the DFA theory. These results suggest that this procedure could help in extracting physical information from the HVSR and thus could lead to an extended use of these measurements to characterize the mechanical properties of the medium.