Zicheng Liu and Michael Cohen
Providing wide angle views is often advantageous to provide context, such as in video conferencing, or when viewing panoramic imagery. Unfortunately, wide angle images often appear distorted. The limited image size on monitors or paper together with standard viewing distances, results in the angle subtended by the image when viewed usually being much smaller than the ﬁeld of view of the camera that originally captured the image. This mismatch is at the root of many perceptual problems. One problem is reﬂected in video conferencing systems where a wide-angle view exaggerates the depth of the room and makes the distant people look extremely small. Other artifacts include the ”swimming-motion” when viewing panoramic images and distortions of objects near the corners of wide angle images. In this paper, we present a class of simple real-time image warping functions that address these perceptual problems. The key idea is to perform local scaling while preserving the global context. We show that the warping functions effectively correct the depth errors in the video conferencing applications resulting in more useful and meaningful conferencing presentations. These warping functions can also be used to remove the swimming motion artifacts when viewing panorama images. In addition, we show a specially designed ﬁve-lens camera to capture and warp wide angle images in real time without sacriﬁcing resolution.