摘要
针对稀疏表示算法横纵采用l^1范数以及l^0范数并不能得到有效的稀疏解的问题,提出了一种将l p(0≤p<1)范数的人脸识别方法。首先通过迭代算法求解l^p范数最小化问题,以此代替传统SRC中的l^1范数来求解编码系数,得到更稀疏和有效的解。为了从稀疏编码系数中捕捉到更多的差分信息,并兼顾残差反映每一类样本的贡献,用系数和与残差之比这一新判别规则来分类测试样本。在AR人脸数据库的实验结果表明,本算法可得到更稀疏有效的解,且可在一定程度上提高识别率,尤其在伪装情况下,有较为明显的提高。
For existing sparse representation classification( SRC) algorithm,l^1 minimization to replace the l^0 minimization does not obtain sufficiently sparse coefficients,a face recognition algorithm combing perfectly the l^p( 0≤p < 1) minimization is proposed. Firstly,solving thelpminimization solution by an iterative algorithm,which replaces the traditional l^1 minimization to obtain the coding coefficients is sparser and more effective. In order to capture more differential information from sparse coding coefficient,and taking into account residual reflects the contribution of each class,using the ratio of the sum of coefficients and the residual to be a new decision rule. The experimental results on AR face database show that the algorithm obtain a sparser solution,and improving the recognition rate to some extent,especially in the guise case,there is a more significant improvement.
出处
《自动化与仪器仪表》
2016年第7期197-199,共3页
Automation & Instrumentation
关键词
人脸识别
稀疏表示
lp最小化
稀疏率
face recognition
sparse representation
lpminimization
sparse rate
