基于Adaboost算法的远距离人脸检测

Long distance face detection based on Adaboost algorithm

为解决在远距离拍摄得到的待识别图像中进行人脸检测难度大、误检率高的问题,提出一种基于Adaboost算法和误报缩减的人脸检测算法。误报缩减是一种结合肤色检测和可变边缘蒙版筛选的模型,肤色检测包含被检测窗口的RGB平均分量,在生成二进制群集图像,边缘蒙版由覆盖二进制集群的椭圆的大小决定,用于在待检测窗口中评价检测对象的轮廓形状,实现误报过滤。实验结果表明,较经典Adaboost算法和Adaboost肤色检测,误报缩减模型在远距离获取的图像中进行人脸检测时具有良好的误报过滤性能。

To decrease the false detection rate of face from the image captured from a long distance,a design method of face detection based on Adaboost algorithm and false report reduction technique was put forward.The false report reduction technique was made up of skin color detection and variable edge board in this method.The average red,green and blue components of the object deteciont window were contained in the skin color detection,followed by the generation of the binary cluster image.The size of edge board was determined by the ellipse covering the binary cluster shape.The edge board was used to filter out false report by evaluating the outline shape of the object in the object detection window.Results show false report rate using this method is lower than that of simple Adaboost algorithm and Adaboost-SkinColor algorithm,and the method is effective for face detection in given images captured at a long distance.