Using Rotated Asymmetric Haar-Like Features for Non-Frontal Face Detection

摘要

The vast applications of face detection system provide the motivation for researchers to find ways of improving the accuracy and performance of the system. One of the challenges in the design of face detection is the high number of false alarm in the testing images. The problem is aggravated for images with pose variations. In this paper, a rotated asymmetric Haar like features using Adaboost cascade classifier is proposed in order to improve face detection with pose variations. The idea of using Asymmetric Haar like features is to highlight the features for non-frontal face images for a more accurate detection. The method is tested using Carnegie Melon University (CMU) database with some added images of veiled women which present problem such as occlusion for the face detection system. The result shows good performance of the method as compared with existing works.