Chapter 5
High level vision tasks, such as object matching and recognition and pose estimation, are performed on features extracted from images. The arrangements of such features are usually irregular and hence the problems fall into categories LP3 and LP4. In this chapter, we present MAP-MRF formulations for solving these problems.
We begin with a study on the problem of object matching and recognition under contextual constraints. An MAP-MRF model is then formulated following the systematic approach summarized in Section 1.5.4. The labeling of a scene in terms of a model(In this chapter, the word "model" is used ti refer ti both mathematical vision models an object models.) object is considered as an MRF. The optimal labeling of the MRF is obtained by using the MAP principle. The matching of different types of features and of multiple objects is discussed. A related issue, MRF parameter estimation for object matching and recognition, will be studied in Chapter 6.
We then derive two MRF models for pose computation. By pose, it means the geometric transformation from one coordinate system to another. In visual matching, the transformation is from the scene (image) to the considered model object (or vice versa). The derived models, the transformation is from a set of object features to a set of image features. They minimize posterior energies derived for the MAP pose estimation, possibly together with an MRF for matching.