An image processing device of the present invention is provided with: a movement amount inference unit (22) that infers a first amount of movement indicating the amount of positional offset of a second image, which is captured after a first image, in directions about three axes with respect to the first image; a region setting unit (23) that, with respect to each of the first image and a first correction image, sets a region to be employed in post-processing, based on feature points extracted from the first image and second image; a post-processing unit (24) that generates, from the first image and first correction image, a first cut-out image and second cut-out image in which the set regions have been cut out, and infers a second amount of movement indicating the amount of positional offset of the second cut-out image with respect to the first cut-out image; and a movement amount combining unit (25) that combines, based on the inferred first amount of movement and second amount of movement, the amounts of movement indicating the amounts of positional offset between the first image and second image.