基于差值区域的实时人脸定位

Real-time face orientation based on difference region

为了解决红外图像人脸跟踪鲁棒性差的问题,提出一种基于差值区域的实时人脸检测和定位方法,该方法主要是利用改进帧间差分法得到一帧差分图像,然后对差分图像进行预处理,确定"类"瞳孔的位置,进而标记出人脸差值区域,对该区域归一化后应用AdaBoost方法进行验证,从而得到人脸的准确定位。研究结果证明:该系统定位方法比直接应用AdaBoost方法有更好的实时性,该方法的各个环节都满足实时性要求,为红外条件下的人脸跟踪系统提供了有效的解决方案。

For addressing the problem associated with the poor robustness of face tracking in infrared images, this study proposes a real-time face detection and orientation method based on the difference region. This method mainly uses the improved frame difference method to obtain a frame difference image. Subsequently, a pretreatment on difference image is conducted to determine the position of ＂similar＂ pupil, and then mark out the regions of difference. After normalizing the difference region, the region is validated using AdaBoost method to obtain an accurate positioning. Experimental study demonstrates that the proposed system has a better performance than that by directly applying the AdaBoost method. All aspects of this proposed method meet the real-time requirements. The proposed method provide an effective solution for face tracking system in infrared images.