Frontal Face Detection In Complicated Background

In 2001, Dr Viola et al. introduced a face detection method using cascade of Haar-like features, which possesses advantages of high-speed and high detection rate and is regarded as one of the most potential method in object detection by the academia.In this paper, a frontal face detection system is proposed by improving the Viola's method. The new system has a high robustness with non-upright faces and is suitable for the face detection in static image with complicated background. The original image and its two rotated images (30° and -30° respectively), are treated as the inputs of the cascade classifier to improve the system's robustness about the non-upright frontal faces. Skin information contained in color images is utilized to decrease the false positive rate and increase the detection speed. The detection results indicate that our attempts are successful. Our detection system achieves a detection rate of 84.5% compared with the 18.5% of the original method which only used a cascade of classifiers in CMU rotated test set. Besides, our system also yields a good result on self-collection test set which contains color images collected from real life and Internet.