【摘要】 眼镜是人脸图像中最常见的遮挡物,它对识别率的影响很大。为了提高戴眼镜人脸图像的识别率,提出了一个从正面人脸图像中提取并摘除眼镜的方法。通过一个自适应二值化的方法,检测了正面人脸图像中眼镜遮挡的部分。经过迭代的误差补偿方法,合成了对应输入图像的不戴眼镜的正面人脸图像。实验结果表明:该方法能有效地合成无眼镜的正面人脸图像。原始戴眼镜人脸图像的识别率是50.1%,合成的无眼镜正面人脸图像的识别率是99.4%。更多还原

【Abstract】 Eyeglasses are the most common occluding object in facial images, which may greatly degrade the recognition performance. This paper describes an algorithm to remove eyeglasses from frontal facial images which first detects the region occluded by the eyeglasses and then synthesizes a natural looking eyeglassless facial image by recursive error compensation using principle component analysis (PCA) reconstruction. The synthesized images have no trace of the eyeglasses or the reflection and shade caused by the eyeglasses. Test results show that the method can effectively synthesize eyeglassless facial images, with the subseguent recognition rate increased from 50.1% to 99.4%.更多还原