Diffracted waves contain a great deal of valuable information about small-scale subsurface structure such as faults, pinch-outs, karsts and fractures, which are closely related to hydrocarbon accumulation and production. Therefore, diffraction separation and imaging with high spatial resolution play an increasingly critical role in seismic exploration. In this study, we apply a method named Geometric-mean Reverse Time Migration (GmRTM) to diffracted waves for imaging only subsurface diffractors based on the difference of the wave phenomena between diffracted and reflected waves. Numerical tests prove advantages of this method on diffraction imaging with higher resolution as well as less artifacts compared to conventional RTM even when we have only small number of receivers. Then we propose a workflow to extract diffraction information using a fully data-driven method, called common reflection surface (CRS), before we apply GmRTM. Application of this workflow shows that GmRTM further improves the quality of the image by combining with diffraction separation technique CRS in the data domain.