An image processing device includes one or more processors configured to: generate a high-resolution combined image by aligning the plurality of images with each other in a high-resolution image space based on an amount of displacement between the plurality of images, and combining the plurality of images; generate at least two low-resolution combined images by generating at least two groups each composed of at least two images by dividing the plurality of images in the time direction, aligning the at least two images in each of the groups with each other in a low-resolution image space based on the amount of displacement, and combining the at least two images through weighted addition; calculate, in each region, a feature quantity pertaining to a correlation between the generated at least two low-resolution combined images; and correct the high-resolution combined image based on the calculated feature quantity.