摘要
光照的变化是影响人脸识别结果的重要因素之一,针对这一问题,提出一种基于加权边缘弱化引导滤波的人脸光照补偿方法。首先为引导滤波损失函数添加一个可区分边缘细节的惩罚项,然后为惩罚项加权,加权系数由正面光照样本的类间平均脸计算得到,最后将滤波后的图像作为自商图中的平滑图,得到光照补偿图像。实验结果表明,该方法弱化了人脸平滑区域由光照造成的边缘细节噪声,且使用光照补偿图像作为人脸识别输入,能有效提高人脸识别准确率,特别在光照大范围变化时,识别准确率提升程度更高。
The variation of illumination is an important factor affecting the face recognition effect.Focusing on this problem,this study proposes a face illumination compensation based on a weighted edge-weakening guided image filter.First,a penalty item,whose edge details can be distinguished,was added into the loss function of the guided image filter.The penalty item was then weighted by a coefficient that was calculated by an inner-class mean face image of positive illumination samples.Finally,the filtered image was used as the smooth image in the self-quotient image to obtain the illumination compensation image.The experimental results showed that the proposed method can weaken the edge noise caused by the illumination in the smooth face area.Moreover,the face recognition rate can be improved using the illumination compensation image as the face recognition input,especially in the case of a large illumination variation.
作者
余拓
陈莹
YU Tuo;CHEN Ying(Key Laboratory of Advanced Process Control for Light Industry Ministry of Education,Jiangnan University,Wuxi 214122,China)
出处
《智能系统学报》
CSCD
北大核心
2018年第3期373-379,共7页
CAAI Transactions on Intelligent Systems
基金
国家自然科学基金项目(61573168)
关键词
人脸识别
光照补偿
光照模型
高斯模糊
引导滤波
岭回归
损失函数
自商图
face recognition
illumination compensation
illumination model
gaussian blur
guided image filter
ridge regression
loss function
self-quotient image