There is provided an image processing device including an edge detection unit configured to detect a boundary point between a first region including a subject to be observed and a second region that does not include the subject, a first estimation unit configured to estimate a first shape as a shape of a boundary between the first region and the second region based on the boundary point, and a second estimation unit configured to estimate a second shape as a shape of a boundary between the first region and the second region based on the boundary point and the estimated first shape.