The Face Detection Based On The Face Landmarks

The face detection is the key technology of the recognition. It is the localization step in the face recognition system. In recent years, with the development of the security access control, man-machine interface recognition and visual surveillance field, face detection began as an independent topic to research and development. It has received wide attention from researchers. Face detection is the process of detecting and extracting human face from static images or videos. The difficulty of the face detection is the face appearance is influenced by itself and external factors. Such as face posture, lightness, and other factors. These factors increased the difficulty of the face detection, but this make the face detection more challenging and significant.We propose a face detection method based on the face landmarks. The innovation of this method is that we deal with the face detection and face landmarks simultaneously and they are in the same cascade framework. Such fusion can improve the detection efficiency and the detection efficiency. Face marked face has a obvious face feature, so it is more likely to be detected. We use the shape index features in the cascade framework, the next stage of the face mark shape depends on the upper stage of the face mark shape, so that the feature has a good geometric invariance. We choose 27 landmarks as the face mark. In each stage of the cascade framework, we will do the face detection and update the 27 face landmarks. That is we will do the face classification and the regression of the face. When we use the cascade face detector to test image, the non-face window will be directly rejected, only the face window enter the next stage of classification and regression. For a test image, we use a sliding window mechanism to scan the picture, the non-face window will be rejected quickly, and the face window is updated 27 face landmarks constantly and eventually detected. The detected face has 27 landmarks, and the face part will be contained in a rectangle.We use classic FDDB face database to do the evaluation of face detection, the evaluation results show that our face detection method is effective. We make a comparison with the widely used Viola-Jones face detector, we use the ROC curve to make the performance evaluation,evaluation results show that our face detection is better than Viola-Jones face detector.