复杂背景下基于人脸的自适应肤色检测

FACE-BASED ADAPTIVE SKIN COLOUR DETECTION IN COMPLEX BACKGROUND

针对传统肤色检测在复杂背景下对类肤色像素的抗干扰能力弱,以及光照对肤色颜色特性影响的问题,提出一种自适应的肤色检测方法。通过Adaboost结合Haar的人脸检测,并利用改进后的二值化提取出包含光照信息的皮肤区域。然后利用改进后的直方图反向投影法,结合肤色样本在YCb Cr空间中Y亮度分量训练得到的最佳阈值,对整幅图像进行肤色检测。实验结果表明,该方法在复杂背景下以及光照的影响均有良好的鲁棒性,在准确率、召回率方面较传统的肤色检测方法有提高。

Traditional skin colour detection has the problems of weak anti-interference capability for similar skin colour pixels in complex background and the influence of illumination on characteristics of skin colour. In light of these,we propose an adaptive skin colour detection approach. First,it extracts the skin area with illumination information by combining Adaboost with Haar face detection and making use of the improved binarisation algorithm. Then,it employs the improved histogram back projection method and combines the Y brightness components of skin colour sample in YCb Cr space to find the optimal threshold through training,and makes skin colour detection on entire image.Experimental result shows that our approach is robust in complex background and under the influence of illumination. Compared with tradition skin colour detection approaches,our approach has the improvement in both precision and recall.