Robust Real-time Face Detection with Skin Color Detection and the Modified Census Transform

We present the combination of an illumination invariant approach to face detection with skin color detection and the false positive suppression used for improving speed and accuracy of the system. First, the non-skin color pixels of the input image are removed based on the mixed skin color model in the YCbCr and RGB chrominance spaces, from which we extract candidate human face regions. Then, face detection with the modified census transform is carried out in the candidate skin color regions where ratio of pixels of skin colour to non-skin color in the sub-windows is above a threshold. To further reduce the computational cost, ratios of pixels of skin color to non-skin color are computed by use of the integral image. Finally, false detection regions are eliminated by using knowledge-based merging and deleting process of homogeneous skin color regions. A face detection system that integrates the skin color detection and false positive suppression method under the proposed framework is built and tested on the Bao image database and two real videos. Experimental results show that our proposed method reduces significantly the false alarms and the processing time while achieves detection rates of more than 99%.