摘要
一般人脸检测率要受到光照的影响,而Gabor小波具有良好的生物视觉特性,对光照不敏感;在用AdaBoost算法训练分类器时,如果人脸在图像中位置相对固定,则分类器的分类能力会增强。为了解决光照问题对检测率的影响,提出人眼定位与AdaBoost Gabor滤波算法相结合的人脸检测方法。采用AdaBoost方法、Hough变换与直接最小二乘法拟合椭圆的方法确定眼睛的区域,并定位瞳孔与眼睑,根据五眼三庭知识得到人脸的区域,最后用AdaBoost Gabor滤波算法训练级联强分类器,并用此级联强分类器进行判别是否为人脸。在Yale、CMU Frontal Face等人脸库上进行实验,实验表明该方法不仅虚警率低,而且人脸检测率高。
Face detection rate is always affected by illumination. However the Gabor wavelet has good biological visual characteristic,and it is insensitive to illumination. In addition,if the position of the face was relatively fixed in images,the ability of the classifiers would be enhanced when classifiers were trained with Adaboost method. In order to increase the face detection rate when face image was affected by illumination,this paper proposed the new face detection method based on the human eye location and AdaBoost Gabor filter. Firstly,it determined eye region with AdaBoost method,Hough transform and direct least squares method. Secondly,it located the pupil and eyelid of the eye. Then,it obtained face region according to the knowledge of the face that its width was five eyes and its high was three court. Finally,it determined face with the cascade strong classifiers which were trained by AdaBoost Gabor filter algorithm. Experiments are executed in Yale,CMU Frontal Face database,and so on. Experiments show that not only false alarm rate with this method is lower,but also face detection rate is higher than other methods.
出处
《计算机应用研究》
CSCD
北大核心
2014年第7期2201-2204,共4页
Application Research of Computers
基金
国家自然科学基金资助项目(61203056)
淮阴工学院基金资助项目(HGB1202)
淮安市工业项目(HAG2013064)
