This doctoral thesis presents a complete framework for automated detection of anomalies in clinical thermal IR images. Within that framework, it addresses two fundamental problems: the automated segmentation of a main object from the background; and the automated extraction of relevant, potentially abnormal Regions of Interest. A first segmentation approach called cued morphological processing of edge maps (CMpEm) is introduced, that uses a minimum amount of a priori information about a region to constrain the morphological processing of edge components. It is shown to outperform existing segmentation methods, especially when dealing with faint regions in challenging conditions. A second segmentation approach based on the classification of edge components is presented that is able to recover missing contours when other methods fail. A third segmentation strategy based on a rule-based fusion and morphological post-processing of segmented contours from different techniques is proposed and shown to improve significantly the performance of the global segmentation task. Furthermore, a new approach for automated extraction of anatomical Regions of Interest in clinical IR images is introduced. It is based on a robust landmark identification algorithm that produces more accurate landmark locations on smooth contours than state-of-the-art algorithms. Various aspects of the anomaly detection framework, including blind extraction of potentially abnormal Regions of Interest, are tested on a collection of thermal IR databases acquired in the course of this thesis, in order to demonstrate the key advantages of the framework.