In view of the existed problems of the existing light field acquisition methods: high hardware cost and big volume in camera array acquisition; to be limited to static light field in single camera ordinal sampling; low spatial resolution of the light field in integral imaging light field acquisition. A method of light field acquisition and reconstruction based on a mask has been studied. A light field acquisition model is set up. Light field imaging process is analyzed in Fourier domain. Utilizing a random mask to randomly encode the light field, which leads to get a coded sensor image. An overcomplete light-field dictionary is learned offline. The coded sensor image is optimized through compressed sensing and the overcomplete light-field dictionary. The original light field is restored.