摘要
为解决在复杂光照条件下的人脸识别问题,提出一种自适应多尺度Retinex(AMSR)和支持向量机(SVM)相结合的人脸识别算法;首先,针对多尺度Retinex(MSR)只能处理光照均匀图像的缺点,提出了AMSR算法,该算法在MSR基础上增加了全局非线性对比度增强方法,使图像的灰度能够根据人脸图像的明暗度进行全局自适应调整,实现了各种光照条件下的人脸图像预处理;然后利用SVM多分类算法对人脸图像进行分类;在人脸库的实验结果证明了AMSR+SVM人脸识别算法的有效性。
In order to solve the problem of face recognition under the complicated lighting conditions, a face recognition algorithm based on the combination of Adaptive Multi--Scale Retinex （AMSR） and Support Vector Machine （SVM） was proposed. At first, aiming at the disadvantage that Multi--Scale Retinex algorithm （MSR） can only deal with the illumination average image, AMSR was proposed. Global nonlinear contrast enhancement was added on MSR in this algorithm, and image＇s grayscale could be adjusted with global and self--adaption according to face image＇s shading value, then face image pretreatment was realized under multiple lighting conditions. Later, face images were classified by SVM multi--class algorithm. The experimental results under face library show the effectiveness of AMSR＋ SVM face recognition algorithm.
出处
《计算机测量与控制》
CSCD
北大核心
2012年第3期823-825,844,共4页
Computer Measurement &Control
基金
国家青年科学基金(61003162)
辽宁省重点实验室(2008s115)
关键词
人脸识别
光照
多尺度RETINEX
自适应
支持向量机
face recognition
illumination
multi--scale retinex
self--adaption
support vector machine
