【作者】 王晓平； 付德刚； 袁春伟；

【Author】 WANG Xiao-ping FU De-gang YUAN Chun-wei (State Key Laboratory of Bioelectronics,Southeast University,Nanjing 210096)

【机构】 东南大学生物电子学国家重点实验室；

【摘要】 针对普通光照条件下的彩色唇读图像,提出了一种结合多种颜色空间及 Fisher 变换的多级唇部检测定位方法。首先在 CbCr 空间建立肤色模型进行人脸检测和定位并由人脸几何特征得到唇部大致区域;然后结合唇色特征对该区域进行 Fisher 变换以使肤、唇色区别明显化;进而根据亮度信息对 Fisher 变换结果进行预处理后再用 Otsu 法自动确定二值化阈值,实验结果表明在唇读图像分割中本文方法较经验阈值法更有效;继而在 rg 空间建立唇色模型对二值化结果进行验证以减少虚检:最后,针对低光照肤色部位可能被误识为唇色的情况使用 HSV 空间中的 H 分量来进行验证。本文中 rg 唇色模型和 H 分量的计算仅在尺寸很小的唇部大致区域进行,其好处是以有限增加的计算量换来了更准确的唇部定位结果。更多还原

【Abstract】 An approach of lip localization for lip-reading under common illumination is proposed here which combines multi-color space and Fisher transformation.First,a skin-color model is constructed for skin detection in CbCr color space so that approximate lip region can be obtained based on geometry features of human face.Then,the chroma difference between lip and skin is enhanced by Fisher transformation.Further,pre-processing of brightness is carded out before segmentation of lip image and the threshold is obtained automatically by Otsu method.The experimental results show that this method is more effective than empirical method in segmentation of lip-reading images.Subsequently,the results are validated by lip-color model in rg color space to avoid wrong detection.Finally,H component in HSV color space is used to reduce error under low illumination.In this paper,the increasement of computation of rg lip-color model and H component is limited due to remarkably shrinking region to be processed.Benefit from this, more accurate lip localization is achieved.更多还原