Adjoint waveform tomography, which is an emerging seismic imaging method for the crust- and global-scale problems, has gained popularity in the past and present decade. This study, for first time, applies adjoint waveform tomography to the large volume of seismic data recorded by the densely spaced, permanent monitoring network that covers the entirety of Japan. We develop a heterogeneous shear-wave velocity model of central Japan that agrees with the geology and lithology. The results reduce the time-frequency phase misfit by 16.4% in the 0.02–0.05 Hz frequency band and 6.7% in the 0.033–0.1 Hz band, respectively. We infer that some velocity anomalies resolved in this work would reflect the subsurface structures such as the volcanic fluids, dehydration of the subducted crust, and sedimentary basin. In addition, dense distributions of deep earthquakes are visible beneath the high-velocity blocks estimated in this study. The results of this study suggest the possibility of imaging large scale heterogeneous subsurface structures using waveform tomography with a densely distributed network of permanent seismometers.