Face Detection plays a major role in Biometrics. Feature selection is a problem of formidable complexity. This paper proposes a novel approach to extract face features for face detection. The LBP features can be extracted faster in a single scan through the raw image and lie in a lower dimensional space, whilst still retaining facial information efficiently. The LBP features are robust to low-resolution images. The dominant local binary pattern (DLBP) is used to extract features accurately. A number of trainable methods are emerging in the empirical practice due to their effectiveness. The proposed method is a trainable system for selecting face features from over-completes dictionaries of image measurements. After the feature selection procedure is completed the SVM classifier is used for face detection. The main advantage of this proposal is that it is trained on a very small training set. The classifier is used to increase the selection accuracy. This is not only advantageous to facilitate the data-gathering stage, but, more importantly, to limit the training time. CBCL frontal faces dataset is used for training and validation.

Keywords: Face features, feature selection, dominant local binary pattern, Support Vector Machine.