Face Detection, Lip Contour Estimation And Tracking For Speaker Identification

There is substantial study based on the topic of face tracking in areas ofcomputer vision and pattern recognition. Basically, all the study can be simplydivided into two categories. One is real time face detection, and the other is thecombination of face detection and face tracking[3~5].Specifically, some face detection algorithms can get the faces in each frametimely, which can directly achieve the goal of face tracking; the other way toimplement face tracking is to detect the face at the start frame first, and then adopt theobject tracking algorithm to follow the face in the other frames.In this project, a combination of basic Haarlike features proposed byViola-Jones an extended set of Haarlike features proposed by Lienhart et al will beresearched and implemented in the face detection phase using matlab and a set ofproposed rules for Adaptive Boosting(Adaboost). Followed by a method of LipLocalization using YCbCr colourspace, lip contour estimation with the Van-Chesemethod of active contours without edges and tracking using Kalman filteralgorithms.The significance of the project is that it can be applied in human-computerinteraction, security and surveillance, video communication and compression,medical imaging, video editing and robot vision applications. This project can alsoserve as groundwork for future research into automatic lip reading systems whichmay make it possible for an application to know what one is saying in a videosimply by reading their lips.