复杂背景的快速人脸检测研究

Study on Fast Face Detection in Complex Background

对于彩色图片,为了能够快速地在复杂背景中检测出多姿态人脸,解决肤色分割后大量粘连的类肤色和肤色区域等问题,在阐述光照和色彩补偿、肤色分割等算法的同时,提出了利用投影法和GLV法相结合的改进算法及利用人的头发来定位人脸的方法。实验结果表明,该方法能够在很短的时间内较为准确地检测出复杂环境中旋转不超过45°的多姿态人脸。因此,具有很好的应用前景。

Human faces need to be detected from the input images with complex background. To deal with the problem of massive adherent resemble skin color and color zones in color images, the algorithms including light and illumination compensation, skin color segmentation are beth discussed. According to these, an improved method inosculating both the projection method and GVL is proposed, which can greatly reduce the number of skin color zones. Finally, a method of locating face by hair on a human head is proposed based on skin color segmentation and connected area segmentation. The experiment demonstrates that this method is competent to detect human faces of various expressions rotated within 45 angles in complex background quickly. This method can apply widely.