The Multi-view Face Detection

Face detection is an important and basic technology which plays a significant role in the research of face. It is not only the first step of face recognition, but also has an independent application foreground. Currently, face detection has been used in many fields such as human computer interface, searches based on content, visions inspection and so on.Through the development of face detection, it has received remarkable achievement. The methods that based on statistics have been used more popular. Paul Viola presented a fast frontal face detection method, which uses Integral Image, and it performs very well. He describes a visual face detection framework that is capable of processing images extremely rapidly while achieving high detection rates. But it is still a hard task to detect all kinds of faces in the image, because faces are various with different condition of lights, expressions and gestures.The paper studies and analyses the fast frontal face detection method. It uses haar-like features to train AdaBoost classifier. Because haar-like features can be computed quickly using Integral Image, it only cost little time for the classifier to detect face. It also uses a method for combining classifiers in a "cascade" which allows background regions of the image to be quickly discarded while spending more computation on promising face-like regions.The paper implements the above method and does many experiments. It gives some discussion and improvement of the original method. First, it improves the training process and short the training time. It also discusses the factors that effect detection rates and false positive rates. In the paper, it presents some new haar-like features and proves that the new features are very efficient in face detection. Finally, it constructs a frontal face detection system.Multi-view face detection is also discussed in the paper. It uses the detector-pyramid architecture consists of several levels from the coarse top level to the fine bottom level. Combined the method of frontal face detection and the detector-pyramid architecture, it designs a multi-view system with 11 classifiers which are able to detect out-of-plane rotations in the range of [-90°, 90°].