Fast facial feature extraction using a deformable shape model with Haar-wavelet based local texture attributes

Wc propose a fast and improved facial feature extraction technique for embedded face-recognition applications. This technique applies to both face alignment and recognition and significantly improves three aspects. First, we introduce the local texture attributes to a statistical face model. A texture attribute characterizes the 2-D local feature structures and is used to guide the model deformation. This provides more robustness and faster convergence than with conventional ASM (active shape model). Second, the local texture attributes are modelled by Haar-wavelets, yielding faster processing and more robustness with respect to low-quality images. Third, we use a gradient-based method for model initialization, which improves the convergence. We have obtained good results dealing with test faces that are quite dissimilar with the faces used for statistical training. The convergence area of our proposed method almost quadruples compared to ASM. The Haar-wavelet transform successfully compensates for the additional cost of using 2-D texture features. The algorithm has also been tested in practice with a Webcam, giving (near) real-time performance and good extraction results.