An image processing apparatus is coupled to a plurality of image capturing devices. The image processing apparatus reduces a noise in an epi-polar image while generating a three-dimensional image from a multi view image. The image processing apparatus divides the multi view image into a flat region and a non-flat region, generates the epi-polar image from the multi view image, replaces an epi-polar line in the epi-polar image corresponding to the flat region with an average pixel value of the multi-view image, and replaces an epi-polar line in the epi-polar image corresponding to the non-flat region with a pixel value of a center-view image obtained from a centrally located image capturing device among the plurality of image capturing devices.