An image processing device includes an image input portion, a motion vector acquisition portion and a distortion component estimation portion. The image input portion inputs a plurality of frame images. The motion vector acquisition portion acquires an amount of parallel movement between two frame images. The distortion component estimation portion performs approximation such that a motion matrix indicating a motion between the frame images consists only of a parallel movement component and a rolling shutter distortion component, uses the amount of parallel movement to calculate, for each of the frame images, a tentative distortion coefficient included in the rolling shutter distortion component and uses the plurality of tentative distortion coefficients calculated to estimate a distortion coefficient.