In comparative image reading of projection images representing an observation target in two three-dimensional images of the same subject photographed at different times, alignment accuracy can be improved between the projection images and artifacts in subtraction images can be suppressed. In each of the three-dimensional images, pixels in a region specified by a cross section perpendicular to an observation direction and a thickness are projected, and the projection images are generated by repeating the procedure while moving a position of the cross section in the observation direction. Alignment processing and subtraction image generation is carried out between the corresponding projection images, and a superposed subtraction image is generated by superposing the generated subtraction images. Consequently, an effect of three-dimensional body movement of the subject at the time of photography can be suppressed than a projection image generated by projecting the entire observation target.