基于RGB颜色空间的红外偏振人脸识别

Infrared Polarized Face Recognition Based on RGB Color Space

热红外成像在人脸识别中具有重要的应用价值,但存在分辨率低、细节不清、边界模糊等不足。通过分析人脸长波红外偏振图像的特征,研究偏振探测技术对人脸热红外成像纹理细节的增强效果;基于高斯差分(DoG)边缘特征图像修正色域通道映射权重,提出一种人脸偏振热像RGB空间融合框架;利用方向梯度直方图(HOG)获取红外偏振人脸特征,提出一种基于支持向量机(SVM)的人脸识别方法。实验结果表明,偏振探测能够增强人脸红外热像的纹理和细节,RGB色域融合可以提高人脸长波红外热像的结构相似性,偏振红外人脸热像的质量指标整体优于普通红外热像;利用偏振红外人脸热像进行人脸识别时,所提框架的精度可达到75.6%。

Thermal infrared imaging has an essential application in face recognition, but it has certain limitations,such as low resolution, unclear details, and fuzzy boundaries. Herein, we describe the enhancement effect of polarization detection technology on the texture details of thermal infrared face imaging by analyzing the characteristics of the long-wave infrared polarization images of human faces. Based on the correction of the difference of Gaussian(DoG) edge feature image’s color gamut channel mapping weights, a RGB space fusion framework for the polarized thermal images of human faces is proposed. We use the histogram of oriented gradients(HOG) to obtain infrared polarization facial features and propose a face recognition method based on support vector machine(SVM). Experimental results show that, first, polarization detection technology can enhance the texture and details of the infrared thermal image of the human face, and that RGB color gamut fusion can improve the structural similarity of the long-wave infrared thermal image of the human face. Second, the overall quality index of polarized infrared thermal images is better than ordinary infrared thermal images. Finally, under the framework of this article,the accuracy for face recognition can reach 75. 6% using the polarized infrared thermal images of the face.