The novel part-based cascaded regression algorithm research combining with pose estimation

摘要

Facial feature location is the key procedure in the process of face image analysis. To increase the localization accuracy and success rate of complex facial images due to lack of shape constraint in robust cascaded pose regression algorithm, the paper had proposed the novel part-based cascaded regression algorithm combining pose estimation. The facial landmarks are divided into several areas and pose estimation has used as shape constraint. To enhance the algorithm's performance, a priori knowledge has been exerted during sampling shape-indexed features to constrain the distance between pixel's pair, and the number of regression has reduced. The experimental results of proposed algorithm and others have demonstrated that the proposed algorithm has outperformed in localization accuracy and achieved better robustness for occlusion and complex facial images, the better localization success rate has been obtained, and the proposed method can realize real-time processing standard.