Local binary patterns and neural network based technique for robust face detection and localization

Abstract

This paper proposes an adaptation of Local Binary Pattern (LBP) operator for the design of the precise face localization technique. Presented approach is divided into three main stages: face detection, detection of eye regions and localization of eye pupils. Introduced optimization principles and the discriminative power of the LBP significantly reduces the number of features used to describe the classes and simplifies the structure of the utilized classifier, namely, artificial neural network. The combination of the reduced dimensionality of the feature space and the presented histogram-based sliding window makes the proposed algorithm well applicable for the task of precise face localization in high resolution images. The parameters of the algorithms are evaluated on a color FERET database.