Robust Active Shape Model using AdaBoosted Histogram Classifiers

Abstract

Active Shape Model (ASM) has been shown to be a powerful tool to aid the interpretation of images, espec ially in fac e alignment. ASM loc al appearanc e model parameter estimation is based on the assumption that residuals between model fit and data hav e a Gaussian distribution. Howev er, in fac e alignment, bec ause of c hanges in illumination, different fac ial ex pressions and obstac les lik e mustac hes and glasses, this assumption may be inac c urate. AdaBoost is widely used in fac e detec tion as a robust c lassific ation method, whic h does not need the Gaussian distribution assumption. I n this paper, we model loc al appearanc es by using AdaBoosted histogram c lassifiers to solv e the robustness problems, whic h hav e prev iously been enc ountered. Ex perimental results demonstrate the robustness of our method to align and loc ate fac ial features.