Locating facial features with an improved active shape model

Abstract

In traditional facial feature localization methods with active shape models, the explicit intensity characteristic of facial features is hardly employed to represent the local structures around them, since these susceptible intensity features are difficult to be extracted. In this work, a novel method is proposed to extract the intensity features based on the logarithmic total variation model. A simple scheme of incorporating the intensity information into the local appearance model is also proposed to validate the effectiveness of the intensity features in the facial feature localization task. Facial feature localization tests are conducted on a wide range of datasets. Experimental results demonstrate the reliability of the proposed method and indicate its superior location accuracy compared to the state of the art method.