Teleconferencing is performed between two telecommunication devices having a display device and a stereoscopic pair of cameras positioned outside opposed sides of the display device at the same level partway along those sides. The separation between the centers of the cameras is in a range having a lower limit of 60 mm and an upper limit of 110 mm to improve the perceived roundness in a displayed stereoscopic image of a head. In captured stereo images that are video images, a head is segmented and the segmented backgrounds are replaced by replacement images that have a lower degree of perceived stereoscopic depth to compensate for non-linear depth perception in the displayed stereo images. Images are shifted vertically to position an eye-line of a detected face at the level of the stereoscopic pair of cameras of the telecommunication device where the images are displayed, improving the naturalness of the displayed image.