Facial Features Localization Based On AdaBoost And Color Information

A great number of researchers worldwide pay more attention on the biometric technologies. Face detection based on AdaBoost and facial features localization based on color information are two important parts which are combined to locate the facial features. In order to further eliminate the background interference, the skin model is built by Gaussian modeling. At last, several contrast experiments are made to compare the method proposed. The result proved that facial features localization has great modeling rate. Therefore the mixed algorithm can be regarded as a great method to locate the facial features.At first, Face detection is realized. All rectangular features are calculated before algorithm implementation. In each loop, the best weak classifier is trained and the image’s weights need to be updated. At last, a strong classifier is generated and can be used to detect the human face.Then, skin modeling based on Gaussian is built. Face likelihood probability map is built in order to reduce the candidate region interference because of the background. Firstly, YCbCr is used for Gaussian modeling. Secondly, the light compensation is conducted based on the reference white light compensation. Then, Gaussian modeling based on the training samples is conducted to produce the skin likelihood probability map. At the same time, global threshold algorithm is used to split the image. At last, two-value image is generated.At last, according to the mixed algorithm, facial feature localization is realized. At first, eye model is established. In YCbCr, eyes region has higher Cb value and less Cr value and the pupil and whites of eyes has contrast of luminance value. Therefore, two eye maps are built and then "and" operation is used to emphasize the common eye. Then, mouth model is also established. In YCbCr, more red pixels and less blue are in mouth compared to other facial features. Therefore, mouth modeling is based on chrominance value. Finally, two-value image generated before is used to eliminate candidate regions from the facial feature modeling image.