Highly accurate extraction of faces and facial parts taking into consideration people with glasses and the specific areas of the face for extracting specific features used in the recognitionof facial expressions

An extraction method for the facial area on a natural background using variable inter-frame subtraction and color information is presented in this paper. We have proposed a human-face extraction technique using the HSV color model and have confirmed the effectiveness of this method experimentally. However, the color black, which is expected to have a lower saturation in the HSV color model, can not be shown accurately. In a conventional system, this feature has some adverse effects on human-face extraction. To cope with this problem, a modified HSV color model was adopted. In the proposed transformation, the saturation is defined as a distance from a chromatic color. This representation has the following advantage: (1) Black is represented as a color with lower saturation. (2) White human skin colors are conspicuous in the conventional system, but they will be emphasized in the proposed system. On the other hand, in terms of moving information, the variable inter-frame subtraction method was adopted for conforming to a natural office environment. However, an adequate frame interval must be evaluated for obtaining sufficient movement information between frames. The most adequate number for obtaining differential images between frames was set in this method. From a computer simulation using 8636 images, 3764 images (97.84%) were correctly extracted. With respect to the extraction of facial parts such as: eyes, mouth, eyebrows, color information and its special mutual relationship were mainly used. Furthermore, the specific areas of a face which are used for feature extraction in order to understand facial expressions were extracted. This process must be executed on all faces taken from an arbitrary angle.