A New Method for Haar-Like Features Weight Adjustment Using Principal Component Analysis for Face Detection
Ramiro Pereira de Magalhães, Cabral Lima · International Conference on Systems · 2014
This paper proposes a new weight assignment method for Haar-like features. The method uses principal component analysis (PCA) over the positive training instances to assign new weights to the features. Together with the method, a particular Haar-like feature that uses statistics extracted from positive training instances is employed. The method and the Haar-like feature were designed to verify if the distribution of points produced from the negative instances in the single rectangle feature space (SRFS) of each Haar-like feature could be modeled as an uniform distribution. Although negative instances may spread themselves in very different and chaotic ways through the SRFS, experiment with the method and the Haar-like feature has shown that the negative instance cannot be properly modeled as an uniform distribution.