|
Face detection is an important topic in the field of computer vision, the earliest as the positioning of links in the face recognition system is proposed. After nearly two decades of development, face detection research has made significant progress, and have a wide range of applications in the field of face recognition, a new generation of human-computer interface, security access and visual monitoring. The face is quite complex details change with non-rigid natural structure of the target, face detection difficult, there are two main aspects, on the one hand, is caused due to internal changes in conditions of the face, such as the details of the face changes relatively abundant, different appearance, such as the face shape, color, etc., as well as different facial expressions, such as eye and mouth, such as opening and closing will result in a change of the details of the human face; caused by the imperfections of the image, such as glasses, hair, as well as due to the occlusion of the face and other accessories, etc.; On the other hand, is caused by external conditions, such as the relative position of the different camera equipment and face the face of multi-state, as well as the impact of external light source changes will result in an image's brightness, contrast changes and shadows change. Face detection process mainly includes feature information extraction and feature classification are two aspects, which feature information extraction is the most important part. Feature extraction common characteristics include the image brightness, shape and texture information. Local Binary Pattern (LBP) is characterized by a very good tool to describe the texture of the image, this paper is part of the picture data is insufficient information (such as low-resolution images), extract features cause certain difficulties, a the new LBP extract operator (ILBP) make up the traditional LBP texture features distinguishing, extracting features of inadequate extraction methods based on traditional LBP defects. And uses a combination of global features and local features, guarantor facial feature information adequacy and stability of the face shape. The experiments show that, compared with the traditional method, this method has higher robustness and practicality. The side using a variety of methods to improve the detection speed and accuracy of human faces: using skin color modeling first extracted skin color area, or to take advantage of the edge detection candidate face region with ILBP operator, and then in the candidate region was extracted from human face characterization face images and features for classification. Finally proposed for face detection in the video stream: the first to detect a motion area in the motion area detection face, which can not only improve the detection efficiency, but also improve the detection speed in the video stream.
|