Real time facial feature localization based on improved active shape model

Abstract

The original active shape model (ASM) has been widely used in facial feature point localization. However, it is neither fast enough nor accurate enough for real time applications. We propose a new search scheme using an active shape model for facial feature extraction. In this scheme, the profile vectors of the original ASM are extended from 2D to 4D. At the same time, a face judgment method is carried out to improve the accuracy of real time fitting for video streams. Experiments using the IMM Face dataset and Frank dataset show that the proposed scheme exhibits better performance than the original ASM in real time video.