Relating images

Starting from a collection of images or a video sequence the first step consists in relating the different images to each other.
This is not a easy problem. A restricted number of corresponding points is sufficient to determine the geometric relationship or *multi-view constraints* between the images. Since not all points are equally suited for matching or tracking (e.g. a pixel in a homogeneous region), the first step consist of selecting a number of interesting points or *feature points*. Some approaches also use other features, such as lines or curves, but these will not be discussed here. Depending on the type of image data (i.e. video or still pictures) the feature points are tracked or matched and a number of potential correspondences are obtained. From these the multi-view constraints can be computed. However, since the correspondence problem is an ill-posed problem, the set of corresponding points can be contaminated with an important number of wrong matches or *outliers*. In this case, a traditional least-squares approach will fail and therefore a robust method is needed. Once the multi-view constraints have been obtained they can be used to guide the search for additional correspondences. These can then be used to further refine the results for the multi-view constraints.

- Feature extraction and matching

- Two view geometry computation

- Three and four view geometry computation