Lip Localization and Viseme Recognition from Video Sequences

Viseme (visual cue) recognition is one of the steps to be followed in building an automated lip-reading system. In order to recognize a viseme, one has to first detect the lips of the speaker from the video sequences and track them to extract the feature vectors for the final recognition. A novel method for liplocalization based on the color models has been proposed. Also, the basic possible lip-shapes depicting the visual-cues have been presented along with their mapping to the corresponding phonemes. In the next level, mapping of the feature vectors from the lip-localization algorithm to the visual cues has been performed.