A video encoding device encodes a multi-viewpoint image as follows. The device encodes an image to be encoded by generating a combined image for a camera (hereinafter, referred to as a first camera) which has captured an image to be encoded, according to an already encoded reference camera image in a viewpoint different from a viewpoint of the first camera and parallax information between the reference camera image and the image to be encoded. A predicted image of a difference image is generated between an input image of the region to be encoded and the corresponding combined image so as to generate a predicted image of the region expressed as a sum of the predicted difference image and the combined image of the region to be encoded. Thus, it is possible to encode a prediction residual expressed by a difference between the predicted image of the region to be encoded and the image to be encoded in the region.