This paper describes a lip localization algorithm within a still image frame for subsequent tracking and audiovisual speech recognition processing. A Gabor filter-based feature space is promoted as a means to localize lips within an image based off of shape. This filtered space is shown to effectively differentiate facial features, including lips, from their backgrounds and to bound the full extent of the lips within a face-classified region of interest. Extensive training and test sets are used to justify design decisions and performance.


Electrical and Computer Engineering



URL: https://digitalcommons.calpoly.edu/eeng_fac/306