Multi-view face detection using Normalized Pixel Difference feature

The initial phase in an automatic facial analysis is face detection, which has been concentrated in the course of recent decades. It is very challenging to recognize a face from an image due to illumination, variations in pose, occlusion, scaling and facial expression. In our work, we propose a methodology to detect a face with arbitrary pose variations. First, a Normalized Pixel Difference (NPD) feature is calculated from the face image. Second, the optimal subset of NPD features and their combinations are refined via Deep Quadratic Tree (DQT). In order to learn the NPD feature based deep quadratic trees, Discrete Adaboost classifier is adopted. The experimental results on Labeled Faces in the Wild (LFW) face dataset demonstrate that our framework performs better than the previous methods in detecting multi-view faces.