Current Issue Cover

张 熠, 张桂林(华中科技大学图像识别及人工智能研究所图像信息处理与智能控制教育部重点实验室, 武汉 430074)

摘 要
提出了一种基于L1总变分模型的对数商图像光照不变人脸识别算法。用L1总变分模型作为低通滤波算子对图像平滑滤波,得到图像光照分量的估计,然后在对数域中定义原图像与其光照分量的商为光照归一化图像,并用该图像作为光照不变量进行人脸识别。基于L1总变分模型的平滑滤波具有较好的边缘保持作用,能有效地消除光晕现象,并且参数设置简单。在YaleB和CMU PIE 人脸图像库上的试验结果表明,该算法能有效地提高人脸识别系统在不同光照条件下的识别率。
An Illumination Invariant Face Recognition Algorithm Based on Total Variation Model

ZHANG Yi, ZHANG Guilin(State Education Commission Key Laboratory for Image Processing and Intelligent Control,Institute for Pattern Recognition and Artificial Intelligence of Huazhong University of Science and Technology, Wuhan 430074)

An illumination invariant face recognition algorithm based on L1 total variation model is proposed. It estimates the illumination from images using L1 total variation model as a low pass filter. Then in log field, the log quotient image is defined as the quotient of the original image and its illumination and is used as normalized illumination invariance for face recognition. TV-L1 based smoothing filter preserves the edges better and can remove halo efficiently. Its parameter selection is also simpler. Experimental results on YaleB and CMU PIE face databases show that the algorithm can effectively improve the face recognition rate under varying lighting conditions.