Before digital recordings became available in the 1970s, the ground motion was recorded using ink on white paper, scratching black-smoked paper, or light on photographic paper. While those analog seismic records offer unique continuous observations from the last century, most of them are now stacked and archived in boxes and potentially exposed to physical decay and permanent loss. To preserve those records and ultimately subject them to modern methods of analysis, it is time-sensitive to scan and digitize them. Here, we worked on a method for automatic digitization of paper seismograms using image processing and machine learning to extract microseismic ground-motion periods and amplitudes. We implemented the method on legacy data recorded at the Royal Observatory of Belgium to extract power spectral densities for major storms during the last century, which are compared with modeled microseisms levels computed using a numerical ocean wave model. This further shows how digitizing analog seismograms does not only preserve the scientific legacy but also makes new research possible by bringing analog data to the digital age.