We introduce a new scheme to study the nature of the central engine in a lensed QSO. The compact emission regions could have different sizes in different optical wavelengths, and our framework permits us to obtain the source size ratios when a microlensing special high-magnification event (e.g., a caustic crossing event, a two-dimensional maximum crossing event and so on) is produced in one of the QSO components. To infer the source size ratios, only cross-correlations between the brightness records in different optical bands are required. While the deconvolution method leads to richer information (1D intrinsic luminosity profiles), the new approach is free of the technical problems with complex inversion procedures. Using simulations related to recent VR data of Q2237+0305A, we discuss how well the scheme is able to determine the visible-to-red ratio q = RV/RR. We conclude that extremely accurate fluxes (with a few μJy uncertainties, or equivalently, a few milli-magnitudes errors) can lead to ˜10% measurements of q. Taking into account the errors in the fluxes of Q2237+0305A from a normal ground-based telescope, ˜10 μJy (˜10 mmag), it must be possible to achieve smaller errors using current superb-telescopes, and thus, an accurate determination of q. To measure the visible-to-red ratio, the light curves must not be contaminated by an intrinsic event or a large high-frequency intrinsic signal, i.e., exceeding the μJy (mmag) level. For an arbitrary lensed QSO, the framework seems to work better with very fast microlensing events. Appendices are only available in electronic form at http://www.edpsciences.org