Regression-Based Sparse Coding for Facial Point Detection

Abstract

Facial point detection in real-world conditions presents large variations in shapes and occlusions due to differences in poses, expressions, use of accessories, which may lead to a large difficultly in locating facial points. In this paper, we propose a regression-based sparse coding method for facial point detection. The method combines the regression-based concept with sparse reconstruction methods to search candidate facial feature points. Specifically, during training, the proposed method learns a group of differential shape dictionaries and Facial point detection in real-world conditions presents large variations in shapes and occlusions due to differences in poses, expressions, use of accessories, which may lead to a large difficultly in locating facial points. In this paper, we propose a regression-based sparse coding method for facial point detection. The method combines the regression-based concept with sparse reconstruction methods to search candidate facial feature points. Specifically, during training, the proposed method learns a group of differential shape dictionaries and local appearance dictionaries. Through system analysis, the results show that our approach outperforms the reference method in terms of detection accuracy.