We have developed a method to estimate surface gravity (log g) from light curves by measuring the granulation background, similar to the "flicker" method by Bastien et al. (2016) but working in the Fourier power spectrum. We calibrated the method using Kepler stars for which asteroseismology has been possible with short-cadence data, demonstrating a precision in log g of about 0.05 dex. We also derived a correction for white noise as a function of Kepler magnitude by measuring white noise directly from observations. We then applied the method to the same sample of long-cadence stars as Bastien et al. We found that about half the stars are too faint for the granulation background to be reliably detected above the white noise. We provide a catalogue of log g values for about 15,000 stars having uncertainties better than 0.5 dex. We used Gaia DR2 parallaxes to validate that granulation is a powerful method to measure log g from light curves. Our method can also be applied to the large number of light curves collected by K2 and TESS.