All of the image analysis techniques reviewed so far are classified as low-level because they can operate directly on the data without any knowledge about the structure of the data or what the data represents (i.e., higher-level knowledge). This is not to say that low-level algorithms always exclude higher-level knowledge, but rather, that they can operate with or without this knowledge (with different levels of performance in each case). It is up to the implementer to decide (1) what types of high-level knowledge the low-level techniques will use and (2) how and when this knowledge will be used. In this section I will review several model-driven image analysis techniques. These techniques either use higher-level information in a fundamentally essential way or use an implicit representation of this higher-level information. I will describe three different model-driven techniques that have been applied successfully in three different application domains: (1) the active contour model, (2) the Kalman filter model, and (3) the Bayesian network model.