A Viseme-based Approach to Labiometrics for Automatic Lipreading

摘要: There are two main approaches to the preprocessing of mouth images in automatic lipreading. The stochastic approach makes wide use learning techniques providing image features poorly interpretable. articulatory aims at measuring as accurately possible anatomical and/or geometrical parameters which can be interpreted phonetic terms. We call this “labiometrics”. Although method here proposed involves processing generally used approach, it is articulatory-oriented indeed; Not only gives some description a shape terms (i.e., visemes), but mostly reliable evaluation labiometric that could not automatically measured on natural lips without prior make-up. Moreover, component our based limited set training images, so its computation cost remains pretty low.