Research On Face Recognition Based On Rough Sets And Transductive Confidence Machine

Face recognition is a branch in Human Identification field, and it is wildly usedin social world because of its uniqueness and stability compared with otheridentification methods. In recent years, following the Probability Theory, Fuzzy Setsand Evidential Theory, Rough set has been developed, which is dealing withuncertainty mathematical problems; and it is also a new soft computing method. Itseffectiveness has been validated in many disciplines and engineering fields. Rough setis one of the current research hot-spots in artificial intelligence field.In view of the analysis of the research progress on face recognition, and combinedwith image-segmentation idea, this dissertation proposes a new feature extractionmethod based on rough sets theory. The rough sets method is used to process thesub-images. Then, the features extracted from sub-images are combined to form thefeature of the whole face image. Based on the ORL face database, BP neural networkis used as the classifier to verify the performance of the feature extraction methods.The results of the experiment are satisfying: in training set, the recognition rate isclose to100%,85%in test set with an average rate close to90%. The experimentalresults show that this feature extraction method works well. On the other hand, inorder to make up for the shortage of lacking quantitative predictions in traditionalclassification methods, transductive confidence machine has been introduced in thisdissertation to further validate the experiment results. What’s more, in identificationstage, transductive confidence can not only provide the prediction results, but alsoguarantee the reliability of those results, which serves as a powerful supplement. Andit is more efficient than the traditional BP neural network in identification. Thus TCMis a practical and applicative classification method.