Tomasi, C. | Kanade, T.
Inferring scene geometry and camera motion from a stream of images is possible in principle, but is an ill-conditioned problem when the objects are distant with respect to their size. An image stream can be represented by the 2F P measurement matrix of the image coordinates of P points tracked through F frames. Based on this observation, the factorization method uses the singular-value decomposition technique to factor the measurement matrix into two matrices which represent object shape and camera rotation respectively. The method can also handle and obtain a full solution from a partially filled-in measurement matrix that may result from occlusions or tracking failures.
The problem of region segmentation includes issues varying from the preprocessing of the image to its semantic interpretation. We first briefly present a model of image understanding to provide a paradigm in which the problem of region segmentation can be discussed. The model enables us to identify three levels of knowledge--signal, physical, and semantic--which play different roles in achieving the goals of region segmentation. This paper emphasizes the importance of exploiting the physical level of knowledge, the bridge between a picture and a scene.