We propose a new approach for describing the geometry information of multiview image representations. Rather than transmitting the raw geometry of the scene, under the form of depth information, we build a graph that represents the connections between corresponding pixels in different views in a multiview image set. The graph starts with the reference image and recursively represents the next levels (i.e., images) by storing the new pixels (those that cannot be derived from the previous image) and their connections to the lower level. The decoder uses these connections to recover the multiple images. In addition to being natural and more easily controlled, the proposed graph-based representation can be compressed more efficiently than depth images. This new representation offers promising perspectives for effective and flexible coding in multiview imaging.