【摘要】 人脸识别中,人脸的归一化处理是至关重要的一个环节,以此为基础可使得人脸的整体分析和识别变得容易和准确。算法首先利用人脸的垂直积分投影曲线获得人脸的左右边界,然后利用水平积分投影曲线获得人眼所处的大致水平位置,进而裁剪出这个包含人眼的大致区域。在此基础上利用Canny算子求出眼睛区域的边缘图像,然后对边缘像素进行聚类,再利用Hough变换的改进方法求出左右眼球中心的精确位置,最后利用中心瞳孔的位置进行人脸的归一化处理。经实验验证,该方法速度快、精确度高,受表情、偏转和光照的影响小。经本方法校正后,后续的识别正确率得到了提高。更多还原

【Abstract】 In face recognition,face normalization is very important as it makes face integral analysis and recognition easier and more accurate.First vertical integral projection curve is used to acquire face’s left and right boundaries,which is followed by the use of horizontal integral projection curve to fix eyes’ approximate horizontal position.Next this approximate area with eyes in it is cut out.By means of Canny operator,edge image in this area is figured out and then clustered.With Hough transform,position of iris is exacted to normalize face.Having been tested by experiments,this method is applied to attain speedy and accurate normalization almost without being subject to facial expressions,rotation and lighting.In conclusion face recognition is more accurate in this way.更多还原