An image processing device includes circuitry configured to: calculate a displacement of each of a plurality of corresponding feature regions between a reference image and a base image; calculate, as a evaluation score, difference value between displacements of two feature regions adjacent to each other in at least one of the up/down direction, the left/right direction, and the oblique 45° direction; determine an abnormal region on the basis of the score; classify other feature regions excluding the abnormal feature region; calculate a projection conversion matrix by using the displacement of the other feature regions and the result of the classification; calculate a degree of alignment of each pixel of the reference image with respect to each pixel of the base image by using the matrix; and generate a combined image by combining the reference image converted based on the degree of alignment with the base image.