An efficient and robust cross-correlation technique for estimation of surface wave Green's function between two locations is applied to a set of ambient noise data recorded by accelerograph networks in the southern Korea. We have estimated 1843 group velocities of Rayleigh waves in the period band of 2--4 seconds from correlation pairs of 91 accelerograph stations. Relatively dense and uniform coverage of the pairs provides us with high-resolution tomographic image of surface-wave group velocities in the region. The resulting group-velocity distribution correlates well with the characteristic features of geological map of the region, displaying low velocities for the sedimentary basins and regions of volcanic and metasedimentary rocks, and high velocities for the regions where granitoid basement rocks outcrop. The short-period information from the accelerograph network can enhance resolution of the subsurface structures at depths within a few kilometers, which may be predominantly responsible for simulation results of strong ground motions.