Color-based Face Detection And Eye Location

The technology of identity authentication that bases on biological characters is developing quickly. Compared with other biometrics technology, face recognition is more direct, friendly and convenient. So, it becomes to be one of the hottest research domains. Face detection is an important pre-processing stage, and the main function is to separate the face from the background. Without a robust, effective and real-time face detection subsystem, the face recognition system can not adapt to real world environment.In color images, the skin-color is important information of face. It is stable, and independent of detail's changes. The skin-color has a good character of clustering in chroma space. According to this character, much background factor can be discarded quickly.Based on the skin-color clustering character, in this dissertation, a face detection algorithm along with an eye locating algorithm are designed and implemented. Three issues were closely investigated,including face detection, illumination compensation and eye locating. The face detection is the basis for the latter two.1. Face detection: firstly, we analyze skin-color's clustering in the YCbCr color space, and establish a Gaussian model based on it. Secondly, a probability image is gained by calculating likelihood though the model. After threshold segmentation, rendering hole, and validating regions, the face regions are finally separated from the background.2. Illumination compensation: experimental results on this face detection method show that the detection rate is reduced greatly because of the hue's and brightness's offset. A global white balance method and composite transformation method are adapted to compensating abnormal images. Experimental results also show that these methods are effective in compensating uneven illumination.3. Eye locating: since the gray level of the eyes'region quite low, the Gaussian model cannot make use of this. The regional skin-color model has gray information which the Gaussian model lacks. Therefore, the dubitable eye's regions are attained by regional skin-color model in the face regions which we have gained. Finally, a quick locating method based on knowledge is presented, and the result is validated by a method based on hue's offset.92 images obtained from Internet and a digital camera were used for the final experiment. And the results indicate that the face detection algorithm and the eye locating algorithm are fast and effective.