Optical wide-field surveys with a high cadence are expected to create a new field of astronomy, so-called “movie astronomy,” in the near future. The amount of data from the observations will be huge, and hence efficient data compression will be indispensable. Here we propose a low-rank matrix approximation with sparse matrix decomposition as a promising solution to reduce the data size effectively while preserving sufficient scientific information. We apply one of the methods to the movie data obtained with the prototype model of the Tomo-e Gozen mounted on the 1.0 m Schmidt telescope of Kiso Observatory. Once full-scale observation with the Tomo-e Gozen commences, it will generate $\sim$30 TB of data per night. We demonstrate that the data are compressed by a factor of about 10 in size without losing transient events like optical short transient point sources and meteors. The intensity of point sources can be recovered from the compressed data. The processing runs sufficiently fast, compared with the expected data-acquisition rate in the actual observing runs.