A foreground extracting section extracts the foreground of each of the first to the N-th frames, and a foreground-accumulated-image configuration section configures a front accumulated image obtained by overlapping the foregrounds of the first to the N-th frames viewed from the future side and a rear accumulated image obtained by overlapping them viewed from the past side. A learning section uses the front accumulated image and the rear accumulated image to obtain prediction coefficients used for predicting the foreground of each frame, and a multiplexer outputs the prediction coefficients, the front accumulated image, and the rear accumulated image as the result of encoding of the foregrounds of the first to the N-th frames.