Face Detection Under Rotation in Image Plane Using Skin Color Model Neural Network and Feature-based Face Model
Hong Zhang · Chinese Journal of Computers · 2002
This paper presents an algorithm for face detection under rotation in image plane. It is a hierarchical approach, which integrates a skin color model, a neural network and a feature-based front face detector. First, the skin color model, which is built for Asian people and Euramerican people, is used to segment face like regions from any input color image. In the face like regions, each 19×19 pixel window at different scales is taken as a face candidate. Then a coarse to fine strategy is designed to acquire accurate face rotation degree with a neural network and by an iris locating operation. According to the acquired rotation degree, each face candidate window is rotated to be upright. Finally, the feature-based front face detector is developed to verify face patterns. Regarding to this detector, this paper proposes a face structure model that is described by three key facial feature points (left eye center point, right eye center point and mouth center point), and provides a method for facial feature point localization. By geometrical relationship of the key facial feature points and texture distribution of face patterns, the front face detector determines whether or not the rotated window is an upright face. Experiments on color images with cluttered scenes show that the presented algorithm is robust for human faces under rotation in image plane, and it is flexible to various lighting conditions and different face sizes.