| Face and gesture recognition are effortless aspects of interaction among humans, but Human Computer Interaction (HCI) remains based upon keyboard and mouse. To facilitate a more fluid interface, we would prefer to enable machines to acquire human-like skills, such as recognizing faces, hand gestures, speech, rather than continuing to demand that humans acquire machine-like skills.; Face and hand gestural recognition are difficult tasks mostly because of the inherent variability of the image formation process in terms of image quality and photometry, geometry, occlusion, change, and disguise. All face and hand gestural recognition systems perform mostly on restricted databases of images in terms of database size, age, gender, race, and/or type of hand gesture, and they further assume well-controlled environments.; Towards that end, we advance in this thesis a novel approach for the design and development of a hybrid learning methodology that is general enough to support both face and hand gesture recognition applications. The hybrid learning methodology consists of explicit connectionist and symbolic modules and implicit fuzzy interfaces as provided by the connectionist networks. The connectionist stage is further defined in terms of ensembles of RBF networks, while the symbolic stage consists of Decision Trees (DT). Specific characteristics of our hybrid learning methodology include (a) query by consensus, as provided by ensembles of networks, for coping with the inherent variability of the image formation and data acquisition process, (b) categorical classification using decision trees, and (c) explanation of the way pattern classification and retrieval is achieved.; Both face and hand gesture recognition lack robust solutions and most of the solutions available so far have never been tested on significant test beds. Experiments conducted on a data set of 900 facial images from the FERET yield an average accuracy of 95.7% for the face recognition task, while experiments performed on a database of 3,000 facial images yield an average accuracy of 96% and 94% for the gender and ethnic classification tasks respectively. Similar experiments conducted for the hand gesture recognition task on a data set 750 images yield an average accuracy of 96.4% respectively. |
