The invention aims at separating a background area of an image using a plurality of cameras by utilizing a method of estimating a background in combination with a stereo method. Data on input images, synchronous with one another, which are obtained by photographing images of a subject from different points of sight using at least two cameras is stored in a frame memory. A background area estimation processing unit performs global image superimposition for the input images, synchronous with one another, the data on which is stored in the frame memory on the assumption that a background area is in the distance and has a flat surface. Then, the background area estimation processing unit estimates the background area based on a degree of local similarity. A distance image generation processing unit performs distance measurement with respect to points other than those in the estimated background area by utilizing a stereo method to generate distance images. A background separated image generation processing unit removes the image of an area in a predetermined distance or more from the input images by referring to resulting distance images to generate a background removed image.