A Discriminative Feature Space for Detecting and Recognizing Faces

Abstract

We introduce a novel discriminative feature space which is efficient not only for face detection but also for recognition. The face representation is based on local binary patterns (LBP) and consists of encoding both local and global facial characteristics into a compact feature histogram. The proposed representation is invariant with respect to monotonic gray scale transformations and can be derived in a single scan through the image. Considering the derived feature space, a second-degree polynomial kernel SVM classifier was trained to detect frontal faces in gray scale images. Experimental results using several complex images show that the proposed approach performs favorably compared to the state-of-the-art methods. Additionally, experiments with detecting and recognizing low-resolution faces from video sequences were carried out, demonstrating that the same facial representation can be efficiently used for both detection and recognition.