摘要
提出了一种适用于复杂背景下的人脸检测新算法。首先对彩色图像进行光线补偿,减小由光源等因素影响引起的色彩偏差;然后对YCrCb色彩格式进行分段线性色彩变换,通过在此基础上建立的肤色模型得到候选人脸区域;最后通过对图像进行总体投影来分离候选人脸区域,再对分离出的区域进行局部投影来确定人脸位置。实验结果表明,该算法可以有效运用于多人脸、不同姿态、不同肤色及复杂背景的情况,对不同的测试图像,检测正确率达到94.7%。
A face detection method based on skin and total-local projection is proposed.First,light compensation is done for a color image to reduce color deviation caused by its light source.Then,piecewise linear color mapping is adopted to the color format YCrCb,on basis of which the skin-color model is set to generate the candidate face region.Finally,the candidate face region is segmented by total projection and a local projection of the regions segmented is employed to determine the face region.The experimental results reveal that this algorithm can be effectively adopted to diverse conditions such as multiple faces,different postures,different color skins and complex backgrounds,with a detection accuracy rate as high as 94.7%.
出处
《红外技术》
CSCD
北大核心
2010年第8期467-470,共4页
Infrared Technology
基金
陕西省自然科学基金项目
编号:2007F05
