A Novel Face Detection and Tracking Method Based on Feature Weighting

We propose a tracking algorithm based on image classification involving online feature weighting. The algorithm uses automatically produced general Haar-like features through feature extraction and feature selection using an online-built object model, and combines Principal Component Analysis (PCA, a generative method) and Fisher ́s Discriminative Analysis (FDA, a discriminative method). That is, we first train the Fisher classifier to distinguish the foreground candidates from background. Then target matching is performed based on similarity with PCA codes of the candidates in feature vectors. The discriminating function of Fisher classifier is a linear combination of the weighted feature values. We also propose a feature discriminative power evaluation equation based on multi-class FDA which gives more discriminative results between the foreground and background. Both the PCA and FDA are online updated to adapt to variation in the images of the tracked object over time, e.g., by noise, occlusion, or a cluttered background. Experimental results show that the proposed method improves detection accuracy when compared with some competitive algorithms. [Chen Y, Park PS. A Novel Face Detection and Tracking Method Based on Feature Weighting. Life Sci J 2014;11(9):23-28]. (ISSN:1097-8135). http://www.lifesciencesite.com. 4