基于HSV与MLPQ的自适应人脸检测算法  

Self-adaptive Face Detection Method Based on HSV and MLPQ

作　　者：周霞[1] 秦磊[1] 王宪[1] 宋书林[1] 

机构地区：[1]江南大学轻工过程先进控制教育部重点实验室,无锡214122

出　　处：《科学技术与工程》2013年第32期9549-9553,9558,共6页Science Technology and Engineering

基　　金：国家自然科学基金(60574051);江苏省产学研联合创新资金-前瞻性联合研究项目(BY2012067)资助

摘　　要：针对复杂背景下彩色图像的人脸检测问题,提出一种基于HSV与多尺度局部量化(MLPQ)的自适应人脸检测算法。首先用HSV空间的肤色模型分割出可能包含人脸的区域。其次用可调的搜索窗口提取可能的人脸区域,并提取其多尺度局部相位量化特征(MLPQ)。最后通过计算人脸候选区域MLPQ直方图与训练样本库中人脸的MLPQ直方图的欧式距离,自适应地进行人脸的确认与定位。在BioID人脸库及彩色FERET人脸库中进行了多组实验,分别取得了92.5%及90.5%的检测率,充分验证了该方法的有效性。Focused on face detection problems of color image under the conditions of the complex background, a self-adaptive face detection method based is proposed on HSV color space and Multiscale Local phase quantization （ MLPQ ）. First, the HSV to do skin color segmentation and located the candidate＇s face area. Then the possible face areas are extracted by adjustable searching window, and the MLPQ features of them are obtained by MLPQ op- erator. Lastly the Euclidean distance is used to compute the MLPQ histogram of possible face area with that of face samples, and self-adaptive locate the face. The experimental results on BioID face database, color FERET face da- tabase can achieve high face detection rate up to 92.5 % ,90.5 % which shows that the proposed method has better effect on improving the detection rate.

关 键 词：HSV颜色空间 多尺度局部相位量化 自适应 人脸检测 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]