Search space optimization and false alarm rejection face detection framework

Abstract

One of the main challenging issues in computer vision is automatic detection and recognition of object classes. In particular, the detection of the class of human faces is a challenging issue, which makes special attention due to the large number of its practical applications, which use face detection as the main and primary step such as face recognition, video surveillance systems, etc. The main aim of face detection is locating human face in images or videos regardless of variations, which are associated to the face detection problem including pose, illumination, and occlusion. The present research distinguishes by two main contributions, which aims to cope with the problem of face detection to locate faces in different poses precisely. The first contribution is the segmentation of face images, based on skin color, which allows discarding the background regions of image quickly. The process aims to decrease the search space and reduce the computation time for feature extraction process. The Second contribution is applying a validation phase in order to reject false alarms. In this phase, the algorithm uses the enhanced local binary pattern and Support Vector Machine (SVM) to extract features of face and classification the features, respectively. In the proposed framework, the intra-class variability of faces is accomplished in a learning module. The learning module used enhanced Haar-like features in order to extract features from human face.